Application of A Zero-latency Whitening Filter to Compact Binary - - PowerPoint PPT Presentation

/24

THE 3RD KAGRA INTERNATIONAL WORKSHOP

Application of A Zero-latency Whitening Filter to Compact Binary Coalescence GW Searches

Leo Tsukada

RESCEU, Univ. of Tokyo The Third KAGRA International Workshop May 21, 2017

1

THE 3RD KAGRA INTERNATIONAL WORKSHOP

/24

This talk is based on …

2

Application of a zero-latency whitening filter to compact binary coalescence gravitational-wave searches

Leo Tsukada,1, 2, ∗ Chad Hanna,3 Cody Messick,3 Drew Keppel,4 Duncan Meacher,3 and Kipp Cannon1, †

1Research Center for the Early Universe (RESCEU), Graduate School of Science,

The University of Tokyo, Tokyo 113-0033, Japan

2Department of Physics, Graduate School of Science,

The University of Tokyo, Tokyo 113-0033, Japan

3The Pennsylvania State University, University Park, Pennsylvania 16802, USA 4

(Dated: April 23, 2017) We examine the performance of a zero-latency whitening filter in a detection pipeline for compact binary coalescence (CBC) gravitational-wave (GW) signals. We find that the filter reproduces sufficiently consistent signal-to-noise ratio (SNR) for both noise and artificial GW signals (called injections) with the results of the original high latency and phase preserving filter. Additionally, we demonstrate that these two filters have a great agreement of squared-chi value, χ2, a discriminator for gravitational wave signals.

Keywords: gravitational waves, compact binary coalescence, whitening filter, low latency

LIGO Document Number “LIGO-P1700094” (In preparation)

THE 3RD KAGRA INTERNATIONAL WORKSHOP

/24

Outline

▸ Background ▸ Introduction ▸ Improvements ▸ Performance tests ▸ Summary

3

OUTLINE

THE 3RD KAGRA INTERNATIONAL WORKSHOP

/24 4

▸ Background ▸ Introduction ▸ Improvements ▸ Performance tests ▸ Summary

Outline

BACKGROUND

THE 3RD KAGRA INTERNATIONAL WORKSHOP

/24 5

▸ Several emissions

๏ Gravitational radiation

Chirp signal

• Short gamma-ray burst

(SGRB) Δt ~ seconds

• Radio afterglow

Δt ~ weeks, years

• Kilonova (optical)

Δt ~ days

BACKGROUND

Metzger & Berger 2012, ApJ 746, 48

NS-NS Coalescence

THE 3RD KAGRA INTERNATIONAL WORKSHOP

/24

Multi-messenger Astronomy

▸ Electromagnetic waves

Gravitational waves Signal association

▸ latency problem

GRB theory : <10s

(X.Li. et al. 2016, ApJ 827, 75)

Pipeline latency : ~30s ← Need to be reduced !

6

BACKGROUND

+

THE 3RD KAGRA INTERNATIONAL WORKSHOP

/24 7

▸ Background ▸ Introduction ▸ Improvements ▸ Performance tests ▸ Summary

Outline

INTRODUCTION

THE 3RD KAGRA INTERNATIONAL WORKSHOP

/24

Inspiral Search Pipeline

▸ Matched filter

8

INTRODUCTION

Whitening transformation

s(t) : data stream h(t) : a template waveform (m1, m2, Deff...) Sn(f) : noise power spectrum density

THE 3RD KAGRA INTERNATIONAL WORKSHOP

/24

Statistics

▸ Signal-to-Noise Ratio : (SNR)

Loudness of the trigger

▸ Chi square :

Discriminator of glitches from chirp signals

9

INTRODUCTION

χ 2

SNR ≡ z σ

≡ σ where σ2 ≡ 4 ∞ |˜ h∗(f)|2 Sn(f) d f

• χ2 ≡

1 σ2/p

p

• i=1

|zi − z/p|2

• where

zi = 4 fi

fi−1

˜ h∗(f)˜ s(f) Sn(f) d f, st ⟨z1⟩ = ⟨z2⟩ · · · = ⟨zp⟩ = ⟨z⟩ p

THE 3RD KAGRA INTERNATIONAL WORKSHOP

/24

Latency

▸ Three bottlenecks

• Data calibration
• Data distribution

๏ Whitening transformation : 16s

10

}

~30s

Essential to improve the whitening filter

INTRODUCTION

THE 3RD KAGRA INTERNATIONAL WORKSHOP

/24

Whitening filter

▸ Flatten the power spectrum ▸ Apply to the both of a template and data

11

• (s(t), h(t)) ∝

∞ ˜ h∗(f)˜ s(f) Sn(f) d f = ∞ ˜ h∗(f)

• Sn(f)

· ˜ s(f)

• Sn(f)

d f

INTRODUCTION

THE 3RD KAGRA INTERNATIONAL WORKSHOP

/24

Whitening filter

▸ Flatten the power spectrum ▸ Apply to the both of a template and data

12

INTRODUCTION

THE 3RD KAGRA INTERNATIONAL WORKSHOP

/24

▸ Discrete Fourier Transform (DFT)

• Applied to 32s blocks every 16s
• Latency of 16s ~ 32s

▸ Frequency-domain whitening

• Conserve the phase of data

Current Algorithm

13

INTRODUCTION

Switch into time-domain processing!

• D533 !(#)

) 33F5F !(#)×&

1536AD 36A36 ()(*) 26F 6 (!×&

+()(*)

• (!×&

• ,!×&

+()

• (#)

) 3D6F5F ,!×&

+()

• (#)×&

)5563 F647 (

Whitening filter

THE 3RD KAGRA INTERNATIONAL WORKSHOP

/24

▸ Discrete Fourier Transform (DFT)

• Applied to 32s blocks every 16s
• Latency of 16s ~ 32s

▸ Frequency-domain whitening

• Conserve the phase of data

Current Algorithm

14

INTRODUCTION

Switch into time-domain processing!

• D533 !(#)

) 33F5F !(#)×&

1536AD 36A36 ()(*) 26F 6 (!×&

+()(*)

• (!×&

• ,!×&

+()

• (#)

) 3D6F5F ,!×&

+()

• (#)×&

)5563 F647 (

THE 3RD KAGRA INTERNATIONAL WORKSHOP

/24 15

▸ Background ▸ Introduction ▸ Improvements ▸ Performance tests ▸ Summary

Outline

IMPROVEMENTS

THE 3RD KAGRA INTERNATIONAL WORKSHOP

/24

Freq.-domain Time-domain Convolution theorem Whitening transformation

Time-domain Processing

16

IMPROVEMENTS

DFT IDFT

!

f[m − n]

n=−∞ ∞

∑

g[n]

s ![m]⋅ 1 Sn[m]

DFT IDFT

!

s[m − n]

n=−∞ ∞

∑

w[n]

F[m] · G[m]

s !( f ) Sn( f ) ⎛ ⎝ ⎜ ⎞ ⎠ ⎟

F[m] ≡ DFT{f} G[m] ≡ DFT{g}

THE 3RD KAGRA INTERNATIONAL WORKSHOP

/24

Freq.-domain Time-domain Convolution theorem Whitening transformation

Time-domain Processing

17

IMPROVEMENTS

DFT IDFT

!

f[m − n]

n=−∞ ∞

∑

g[n]

s ![m]⋅ 1 Sn[m]

DFT IDFT

!

s[m − n]

n=−∞ ∞

∑

w[n]

F[m] · G[m]

Finite Impulse Response (FIR)

s !( f ) Sn( f ) ⎛ ⎝ ⎜ ⎞ ⎠ ⎟

F[m] ≡ DFT{f} G[m] ≡ DFT{g}

THE 3RD KAGRA INTERNATIONAL WORKSHOP

/24

Linear-phase FIR filter

18

IMPROVEMENTS

Latency 16s Amplitude response Linear-phase filter

THE 3RD KAGRA INTERNATIONAL WORKSHOP

/24

Minimum-phase FIR Filter

19

IMPROVEMENTS

Zero latency filter !

Minimum-phase filter Linear-phase filter

THE 3RD KAGRA INTERNATIONAL WORKSHOP

/24 20

▸ Background ▸ Introduction ▸ Whitening filter ▸ Improvements ▸ Performance tests ▸ Summary

Outline

PERFORMANCE TESTS

THE 3RD KAGRA INTERNATIONAL WORKSHOP

/24

Statistical tests

21 Auto-correlation Amplitude histogram

Good whitening quality !

PERFORMANCE TESTS

THE 3RD KAGRA INTERNATIONAL WORKSHOP

/24

New Whitener vs Old Whitener

▸ Data

• 45,056 s during S5 at the Hanford LIGO detector

▸ Noise-based test

• Trigger-trigger (only noise) association
• and comparison

▸ Injection-based test

• Simulated chirp signals injected every 31.4s
• Injection-injection association
• and comparison

22

PERFORMANCE TESTS

SNR

χ 2

SNR

χ 2

THE 3RD KAGRA INTERNATIONAL WORKSHOP

/24

Noise-based test

23

PERFORMANCE TESTS

THE 3RD KAGRA INTERNATIONAL WORKSHOP

/24

Injection-based test

24

PERFORMANCE TESTS

Agreement between the old and new whiteners !

THE 3RD KAGRA INTERNATIONAL WORKSHOP

/24

Improved latency

▸ Three bottlenecks

• Data calibration
• Data distribution

๏ Whitening transformation : 16s

25

}

Significant role in the whole latency reduction

INTRODUCTION

0s

18s

THE 3RD KAGRA INTERNATIONAL WORKSHOP

/24 26

▸ Background ▸ Introduction ▸ Whitening filter ▸ Improvements ▸ Performance tests ▸ Summary

Outline

SUMMARY

THE 3RD KAGRA INTERNATIONAL WORKSHOP

/24

Summary

▸ Dawn of multi-messenger astronomy

• Signal association from NS-NS merger

▸ Latency problem

• Need to be ~ 10s

▸ FIR whitening transformation

• Zero latency whitening by minimum-phase filter
• Good agreement between the old and new whiteners

▸ Improved latency of 18s

• The new whitener will be implemented soon.

27

SUMMARY

THANK YOU FOR LISTENING !