---
title: 5-Vector Resampling in GW Searches
url: https://www.emergentmind.com/topics/5-vector-resampling
type: topic
---

# 5-Vector Resampling in GW Searches

Searching arXiv for the cited five-vector resampling papers and closely related work.
{"query":"five-vector resampling continuous wave Scorpius X-1 arXiv", "max_results": 10}
{"query":"A directed continuous-wave search from Scorpius X-1 with the five-vector resampling technique", "max_results": 5}
{"query":"2503.07863", "max_results": 5}
5-vector resampling is a data-analysis technique for directed and narrowband searches for continuous gravitational waves from neutron stars in binary systems. It combines time-domain resampling, which removes Doppler and relativistic phase modulations, with the five-vector formalism, which represents the residual sidereal amplitude modulation in five Fourier components. In the formulation applied to Scorpius X-1 and to directed searches from neutron stars in binary systems, the method is designed to recover the coherent signal-to-noise ratio otherwise lost when orbital motion spreads signal power across many frequency bins, while maintaining an affordable computational cost for searches over uncertain orbital parameters [2503.07863], [2505.19041].

## 1. Physical setting and search motivation

Continuous gravitational-wave signals are treated as nearly monochromatic in the neutron star rest frame, with phase
$$
\Phi^{\rm NS}(\tau)=\phi_0+2\pi f_{\rm GW}(\tau-\tau_0).
$$
For sources in binaries, the phase observed at an Earth-based detector is modulated both by the Earth’s orbital and rotational motion and by the neutron star’s orbital motion around the binary barycentre. The latter is typically described by five Keplerian parameters, and inaccurate treatment of those parameters causes the signal to be smeared over multiple FFT bins, reducing the signal-to-noise ratio and potentially hindering detection [2505.19041].

In the detector frame, the phase can be written as
$$
\Phi^{\rm det}(t_{\rm arr})=\phi_0+2\pi f_{\rm GW}\bigl[t_{\rm arr}+\Delta\tau(t_{\rm arr})-\tau_0\bigr],
$$
where $\Delta\tau(t)$ includes Earth-barycentric corrections and the binary delay. In the low-eccentricity description emphasized in the method summary, the binary Rømer delay depends on the five Keplerian parameters $\{P,a_{\rm p},e,\omega,t_{\rm p}\}$; equivalently, for $e\ll 1$, one may use Laplace–Lagrange parameters $\{\Omega,a_{\rm p},\kappa,\eta,t_{\rm asc}\}$ with $\Omega=2\pi/P$, $\kappa=e\cos\omega$, $\eta=e\sin\omega$, and $t_{\rm asc}\equiv t_{\rm p}+\omega/\Omega$ [2505.19041].

The search motivation is astrophysical as well as algorithmic. The Advanced LIGO-Virgo-KAGRA detectors operate in bands containing more than half of the known pulsars in the Galaxy existing in binary systems, and the method was developed specifically to conduct thorough directed and narrowband continuous-wave searches over such targets at affordable computational cost [2503.07863].

## 2. Phase demodulation through resampled time

The central step of 5-vector resampling is the construction of a new time coordinate that unwinds the detector-frame phase modulation. In the Scorpius X-1 implementation, the detector phase is written
$$
\Phi^{\rm det}(t_{\rm arr})=\phi_0+2\pi f_{\rm GW}\Bigl[t_{\rm arr}+\frac{\mathbf x(t_{\rm arr})\!\cdot\!\hat r}{c}-\frac{R(t_{\rm arr})}{c}+\Delta_{\rm Ein}-\Delta_{\rm S}-\tau_0\Bigr],
$$
with $R(t)$ the projected radial binary delay and $\mathbf x(t_{\rm arr})\!\cdot\!\hat r/c$ the geometric barycentric term [2503.07863].

The resampled time variable is then defined as
$$
t' = t_{\rm arr}
+ \frac{\mathbf x(t_{\rm arr})\!\cdot\!\hat r}{c}
- \frac{R(t_{\rm arr})}{c}
+ \Delta_{\rm Ein}
- \Delta_{\rm S},
$$
or, equivalently in the compact notation of the broader method description,
$$
t' \equiv t_{\rm arr}+\Delta\tau(t_{\rm arr}).
$$
By construction, the phase in the new coordinate is strictly linear,
$$
\Phi^{\rm det}(t')=\phi_0+2\pi f_{\rm GW}(t'-\tau_0),
$$
so the astrophysical signal becomes nearly strictly monochromatic in the resampled time series [2503.07863], [2505.19041].

Operationally, the original uniformly sampled detector data $x(t_{\rm arr})$ are resampled onto the irregular grid $\{t'_n\}$ and represented as a new time series $x'(t')$. The method summary explicitly notes that no interpolation of frequency is needed because the resampling is performed in the time domain [2503.07863]. This is the defining distinction of the technique: the orbital and barycentric corrections are absorbed into the time coordinate rather than re-applied separately to every trial frequency.

## 3. Five-vector formalism and matched filtering

After phase demodulation, the signal retains the Earth’s sidereal amplitude modulation through the detector antenna patterns. In the Scorpius X-1 formulation,
$$
h(t')=H_0\bigl[H_+\,A_+(t')+H_\times\,A_\times(t')\bigr]
\exp\bigl\{i[2\pi f_{\rm GW}t'+\phi_0]\bigr\},
$$
where $A_{+,\times}(t')$ are known sidereal antenna patterns [2503.07863].

The five-vector is constructed by taking Fourier components at the five sidereal sidebands. For a real time series $g(t')$,
$$
G_k(f)=\int_0^{T_{\rm obs}} g(t')\,e^{-2\pi i (f-k f_{\rm sd}) t'}\,dt', \qquad k=-2,\dots,2,
$$
and
$$
\mathbf G(f)=\bigl[G_{-2}(f),G_{-1}(f),G_0(f),G_{+1}(f),G_{+2}(f)\bigr]^{\rm T}.
$$
The Earth sidereal frequency is given as $f_{\rm sd}\approx1.16\times10^{-5}\,\mathrm{Hz}$, so the signal is represented by five components at $f_{\rm GW}+k f_{\rm sd}$ with $k=0,\pm1,\pm2$ [2503.07863], [2505.19041].

Under the signal model, the data five-vector is
$$
\mathbf X=H_0 e^{i\phi_0}\bigl(H_+\,\mathbf A^+ + H_\times\,\mathbf A^\times\bigr)+\mathbf N,
$$
with $\mathbf A^{+,\times}$ the five-vectors of the sidereal templates and $\mathbf N$ the noise vector [2503.07863]. The broader formulation writes this as a $2\times5$ linear model,
$$
X = A\cdot [h_+,h_\times]^{\rm T} + N,
$$
from which a generalized-likelihood-ratio statistic can be formed [2505.19041].

In the Scorpius X-1 pipeline, polarization estimators are computed as
$$
\hat h^{+,\times}(f)=\frac{\mathbf X(f)\cdot \mathbf A^{+,\times}}{\|\mathbf A^{+,\times}\|^2},
$$
and the detection statistic is
$$
{\cal S}(f)=\|\mathbf A^+\|^4\,|\hat h^+|^2+\|\mathbf A^\times\|^4\,|\hat h^\times|^2.
$$
The companion methodological summary presents the corresponding optimal statistic as $2\mathcal F=X^\dagger W X$, where $W$ is the inverse noise-covariance matrix in five-vector space [2505.19041]. Taken together, these descriptions show that the method pairs explicit binary demodulation with a coherent sidereal matched filter.

## 4. Pipeline realization in directed and narrowband searches

The high-level implementation described for the Scorpius X-1 search begins with construction of the SFDB (Short-FFT Database). The full band is divided into 1-Hz sub-bands, short FFTs are formed, and calibrated strain is recorded. For each point in the search grid over orbital-parameter offsets $\{\Delta P,\Delta a_{\rm p},\Delta e,\Delta\omega,\Delta t_{\rm p}\}$, the mapping $t_{\rm arr}\to t'$ is computed and the original time series is interpolated to obtain $x'(t')$ on a uniform grid. This time-domain interpolation is the most expensive step in CPU terms [2503.07863].

The broader methodological description expresses the same logic in a band-limited form. Starting from calibrated evenly sampled data $s(t_{\rm arr})$, one computes $\Delta\tau(t_{\rm arr})$, forms the resampled timestamps $t'$, chooses a uniform grid in $t'$ with $\Delta t=1/(2f_{\max})$ to satisfy Nyquist sampling to $f_{\max}$, and interpolates with a high-order kernel. The resampled series may then be heterodyned by $e^{-2\pi i f_0 t'}$, low-pass filtered over a narrow bandwidth $B$, and downsampled before an FFT is taken and the five sideband bins are extracted [2505.19041].

Candidate selection is not based solely on the raw detection statistic. In the directed Scorpius X-1 pipeline, the noise-only distribution $f(\mathcal S)$ is estimated, a threshold $\mathcal S^*$ is set by fixing the false-alarm rate through the look–elsewhere factor, and the statistic is normalized to
$$
\mathcal S'=(\mathcal S-\mu)/\sigma
$$
to provide robustness against band-by-band variations and instrumental lines. Candidates with $\mathcal S'>\mathcal S'_*$ are retained. They are then subjected to an internal same-detector coincidence veto requiring at least $n_{\rm coin}^{\rm int}$ out of the 9 possible sidereal-sideband peaks to exceed threshold, followed by an inter-detector coincidence condition requiring appearance in both Hanford and Livingston within $1/T_{\rm obs}$ of the expected sidereal sidebands [2503.07863].

## 5. Sensitivity recovery and computational scaling

The method is motivated by a specific sensitivity loss mechanism. Without demodulation, orbital motion spreads the signal over $N_{\rm bins}\sim \Delta f_{\rm orb}\,T_{\rm obs}$ bins, where the orbital-Doppler bandwidth is given as $\Delta f_{\rm orb}\sim (2\pi f_{\rm GW} a_{\rm p})/P$. The coherent SNR is therefore reduced by approximately $1/\sqrt{N_{\rm bins}}$. After exact resampling, the signal power is refocused into the five sidereal bins, and the SNR gain is summarized as
$$
\frac{\mathrm{SNR}_{\rm demod}}{\mathrm{SNR}_{\rm undemod}}
\simeq \sqrt{N_{\rm bins}}
\simeq \sqrt{\frac{2\pi f_{\rm GW} a_{\rm p} T_{\rm obs}}{P}}.
$$
The Scorpius X-1 summary states the same point qualitatively: longer coherence time raises SNR as $\propto \sqrt{T_{\rm obs}}$, and the five-vector filter concentrates all signal power into nine peaks, yielding a matched-filter boost [2505.19041], [2503.07863].

The computational scaling is likewise explicit. For one 1-Hz band and one Keplerian template in one detector, the total CPU cost is approximated by
$$
\mathcal C_{1\,\rm Hz}
\approx c_1 T_{\rm obs}
+ c_{\rm res} T_{\rm obs}
+ c_{\rm DS} T_{\rm obs}
+ c_{\rm coinc} T_{\rm obs},
$$
with $c_1\simeq 5.5\,\mathrm{s/day}$ for band extraction, $c_{\rm res}\simeq 68\,\mathrm{s/day}$ for resampling, $c_{\rm DS}\simeq 0.52\,\mathrm{s/day}$ for the five-vector filter, and $c_{\rm coinc}\simeq 8.6\,\mathrm{ms/day}$ for coincidence checking. For $N_f$ 1-Hz bands, $N_p$ orbital templates, and $N_{\rm det}$ detectors,
$$
\mathcal C_{\rm tot}=N_f N_p N_{\rm det}\,(c_1+c_{\rm res}+c_{\rm DS}+c_{\rm coinc})\,T_{\rm obs}.
$$
The dominant cost is therefore the time-domain resampling step [2503.07863].

A crucial optimization is that the time-resampling step is independent of frequency. A single resampling covers the entire 1 Hz band, and the same resampled time series can be reused for all sub-bins. The method summary identifies this as a major saving relative to recalculating Doppler corrections for each frequency and notes that this feature makes the approach ideally suited to narrowband searches of unknown spin frequency [2503.07863]. A plausible implication is that the method’s practical niche is precisely the regime in which orbital-parameter uncertainty is significant but the source sky position is already well constrained.

## 6. Application to Scorpius X-1 and methodological limits

The first application described in the supplied material is a directed search for continuous waves from Scorpius X-1 using publicly available data from the third observing run of Advanced LIGO-Virgo-KAGRA. The search used the full O3 run from Hanford and Livingston over a frequency range of 10–1000 Hz and tested three ephemeris configurations: the nominal Scorpius X-1 parameters and two edge-of-uncertainty offsets in $\{\pm\Delta P,\pm\Delta a_{\rm p},\pm\Delta e,\pm\Delta\omega,\pm\Delta t_{\rm p}\}$ [2503.07863], [2505.19041].

No statistically significant continuous-wave signal was claimed, and no candidates survived the follow-up chain. The search therefore set 95% confidence-level upper limits in selected frequency bands and orbital-parameter ranges while also evaluating overall sensitivity [2503.07863]. The method summary reports that continuous-wave signals down to strain amplitudes $h_0\sim 5\times10^{-26}$ can be recovered at 95% confidence in O3 data for Scorpius X-1, while the broader summary gives a best 95% confidence-level upper limit of $h_0^{95\%}\simeq 5.6\times10^{-26}$ at $f\approx 229.5\,\mathrm{Hz}$ and, more specifically, a best value $h_0\approx 5.64\times10^{-26}$ at $229.5\,\mathrm{Hz}$ in LIGO Livingston [2503.07863], [2505.19041].

Several practical limitations are emphasized. The robustness of the method depends on covering uncertain orbital parameters within maximum offsets that still produce a coherent peak. For Scorpius X-1, the electromagnetic uncertainties exceed those offsets, so three searches—null, “+”, and “−” offsets—were used to cover the ephemeris. Known spectral lines and their Doppler-broadened replicas must be zeroed before fitting the noise tail; otherwise the threshold can be overestimated by up to an order of magnitude. The method is therefore not a generic black-box demodulator: its performance depends on orbital ephemeris control, line handling, and a coincidence framework tailored to the sidereal-sideband structure [2503.07863].

A recurring misconception is that resampling renders all detector modulation irrelevant. The formalism shows otherwise. Resampling removes the phase modulation associated with barycentric and binary delays, but the Earth’s sidereal amplitude modulation remains and is intentionally retained as the structure exploited by the five-vector matched filter [2503.07863]. That separation between phase demodulation and sidereal filtering is the defining feature of 5-vector resampling.

Source: https://www.emergentmind.com/topics/5-vector-resampling