timing methods in x-ray astronomyiac.es/winterschool/2015/media/lectures/belloni_timing2.pdf · if...
TRANSCRIPT
![Page 1: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/1.jpg)
TIMING METHODSIN X-RAY ASTRONOMY
Tomaso Belloni (INAF - Osservatorio Astronomico di Brera)
![Page 2: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/2.jpg)
POISSON NOISE EFFECTS
• Counting detector
• Counting noise
• Background negligible
• Independent arrival times
• Exponential waiting time between photons
3 6 9
![Page 3: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/3.jpg)
POWER SPECTRUM NORMALIZATION
•With this choice, noise power a χ2 with 2 d.o.f.
•Most noises do
• Average power is 2. I can calculate statistics
•Noise & signal independent:
•Not always so... (count rate!)
•More complex: deadtime
P =2
Nphot|a|2
Leahy Norm.
![Page 4: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/4.jpg)
POWER OF POWER SPECTRUM
![Page 5: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/5.jpg)
NOISY NOISE✤ Power spectrum of noise
is very noisy!
✤ Increasing length or Δt not useful
✤ Two ways out:
✤ a) Frequency rebinning by M
✤ b) Time slicing by W and averaging powers
M=10
2 with 2MW dof2MW distribution scaled by MW
Mean: 2Standard dev:
![Page 6: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/6.jpg)
FULL POWER SPECTRUM✤ RXTE light curve
✤ t = 1/16 seconds
✤ T = 3325 seconds
✤ Something can be seen by eye in the light curve
✤ Full power spectrum
✤ High-power signal, no coherent peak
![Page 7: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/7.jpg)
LOG SPACE AND REBINNING✤ Log-log plot more appropriate
for all frequencies
✤ Errors are 100%
✤ Frequency rebinning (M)
✤ Log-rebinning:
✤ Error bars, better shape
✤ Poisson level below scale
![Page 8: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/8.jpg)
NORMALIZATION✤ Leahy normalization very useful for
statistics
✤ Power ∝ square intensity
✤ Remove it by dividing by square intensity: rms (Belloni) normalization
✤ Caveat: from Leahy to rms2
✤ Meaning: squared rms per decade
✤ Root of integral gives fractional rms
![Page 9: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/9.jpg)
A NOTE ABOUT REBINNING
✤ Coherent peak: narrow power distribution - least rebinning - the longer the observation span, the better
✤ Broad peak: broad power distribution - rebinning helps - length of observation not crucial
✤ Very important for maximizing sensitivity
![Page 10: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/10.jpg)
W: WELCH POWER SPECTRUM
✤ If signal stationary
✤ Slice the signal
✤ Power spectrum of slices
✤ Add the W slices
✤ Sliding slices are also possible (statistics?)
✤ Windowing is also possible
![Page 11: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/11.jpg)
TIME-FREQUENCY ANALYSIS✤ If signal is not stationary
✤ No average of power spectra
✤ Image: time-frequency-power
✤ Uncertainty principle
![Page 12: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/12.jpg)
THE UNCERTAINTY PRINCIPLE
• You cannot beat it
• It’s a big limitation
Time
Freq
uenc
y
T 2 = �2t =
Z(t� < t >)2|s(t)|2dt
B2 = �2! =
Z(!� < ! >)2|S(!)|2d!
TB � 1
2
�t
�!
DurationBandwidth
![Page 13: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/13.jpg)
THE EASY WAY OUT• Spectrogram (from short-term Fourier Transform)
• Sliding window to select time (window can be chosen)
•Obtain a time-frequency image
st(⌧) = s(⌧)h(⌧ � t)
P (t,!) =
����1p(2⇡)
Re(�i!⌧)s(⌧)h(⌧ � t)d⌧
����2
![Page 14: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/14.jpg)
AN EXAMPLE•Quasi-Periodic Oscillation T = 128 s
![Page 15: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/15.jpg)
AN EXAMPLE•Quasi-Periodic Oscillation T = 4 s
![Page 16: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/16.jpg)
NON-OVERLAPPING• Sliding window to select time st(⌧) = s(⌧)h(⌧ � t)
t = ⌧
![Page 17: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/17.jpg)
NON-OVERLAPPING• Sliding window to select time st(⌧) = s(⌧)h(⌧ � t)
t < ⌧
![Page 18: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/18.jpg)
SHIFT ‘N’ ADD TECHNIQUE✤ Used for twin high-frequency peaks
✤ You see one, not the other
✤ The one you see moves
✤ Correct for the movement, align the spectra in an additive way
✤ More complex: multiplicative technique (tricky to implement)
![Page 19: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/19.jpg)
LINEAR SHIFT AND ADD• Good to recover features at a constant distance in ν
![Page 20: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/20.jpg)
INSTRUMENTAL DEAD TIME✤ After a photon, dead time
✤ Introduces correlations between photons (no Poisson!)
✤ It must be as small as possible and well-known and modeled
✤ Two types of dead time:
✤ Paralyzable
✤ Non-paralyzable
Every incident event causes a dead time td
even if it’s not detected
Only a detected event causes a dead time td
![Page 21: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/21.jpg)
PARALYZABLE DEAD TIME
✤ If incident rate rin is very high, no detected counts at all!
✤ Detected rate:
✤ In RXTE/PCA, for binning time
![Page 22: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/22.jpg)
PARALYZABLE DEAD TIME
rin = 20 kcts/sr0 = 16.385 kcts/std = tb = 10 µsN = 1024
![Page 23: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/23.jpg)
NON-PARALYZABLE DEAD TIME
✤ If incident rate rin is very high, one count every td
✤ Detected rate:
✤ Formula is even more complicated, result is similar
✤ Depression of noise level @ low frequencies (correlation)
✤ Peak @ td (quasi-periodicity)
![Page 24: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/24.jpg)
PARALYZABLE DEAD TIME: SCO X-1
r0 = 105 cts/std = 10 µs
Source noise
kHz QPOs
![Page 25: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/25.jpg)
FITTING POWER SPECTRA✤ Fit with typical minimization (χ2)
✤ Rebinning is important for χ2
✤ Error estimation vs. significance
✤ Limit in power an NOT rms
✤ Coherent peaks: distribution of powers and number of trials
![Page 26: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/26.jpg)
NUMBER OF TRIALS✤ Important statistical concept
✤ Should be done correctly, but if P is small can be approximated
✤ IMPORTANT: how to estimate Ntrials
✤ For Power Spectra: number of independent frequencies
![Page 27: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/27.jpg)
CONTINUUM COMPONENTS
✤ Very important for accreting sources
✤ Slope is limited by the window
✤ Window overflow
✤ Γ=-2 is the steepest value
✤ If an issue (pulsar noise): exotic methods
Power law Γ=-2
Power law Γ=-1
Lorentzian
Sinc
![Page 28: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/28.jpg)
MAIN TYPES OF SIGNALS
✤ Coherent pulsation
✤ Broad-band noise
✤ Broad peak (QPO)
✤ “Peaked-noise”
![Page 29: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/29.jpg)
MAIN TYPES OF SIGNALS
✤ Coherent pulsation
✤ Broad-band noise
✤ Broad peak (QPO)
✤ “Peaked-noise”
![Page 30: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/30.jpg)
MAIN TYPES OF SIGNALS
✤ Coherent pulsation
✤ Broad-band noise
✤ Broad peak (QPO)
✤ “Peaked-noise”
![Page 31: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/31.jpg)
MAIN TYPES OF SIGNALS
✤ Coherent pulsation
✤ Broad-band noise
✤ Broad peak (QPO)
✤ “Peaked-noise”
![Page 32: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/32.jpg)
MAIN TYPES OF SIGNALS
✤ Coherent pulsation
✤ Broad-band noise
✤ Broad peak (QPO)
✤ “Peaked-noise”
![Page 33: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/33.jpg)
MAIN TYPES OF SIGNALS
✤ Coherent pulsation
✤ Broad-band noise
✤ Broad peak (QPO)
✤ “Peaked-noise”
![Page 34: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/34.jpg)
A WORD ON REPRESENTATION
![Page 35: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/35.jpg)
A WORD ON REPRESENTATION
![Page 36: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/36.jpg)
THE LORENTZIAN (ZERO-CENTERED)
✤ Power spectrum of a one-sided exponential
✤ Good for modeling broad-band noise components (flat-top)
![Page 37: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/37.jpg)
THE LORENTZIAN✤ Centroid of Lorentzian not at
zero
✤ Good for modeling Quasi-Periodic Peaks
![Page 38: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/38.jpg)
THE QUALITY FACTOR Q
✤ To quantify the coherence of a component
Q=10, 5, 2.5, 1.25 0.625
![Page 39: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/39.jpg)
Q=0: THE PEAK WITHOUT QUALITY
✤ Here ν0 = 0, equal N
✤ Notice position of the break
✤ Factor of two higher Δ
![Page 40: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/40.jpg)
BETTER REPRESENTATION
✤ In νPν the effect is the same
✤ Better value is Δ/2
✤ But... how do I treat thingshomogeneously and howdo I treat peaked noise?
![Page 41: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/41.jpg)
CHARACTERISTIC FREQUENCY
✤ We can use the peak in νPν
![Page 42: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/42.jpg)
LORENTZIAN DECOMPOSITION
✤ With these tools we can fit power spectra
![Page 43: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/43.jpg)
NO PHYSICAL BACKING (YET)
✤ Power spectrum of a damped oscillator
✤ Also called Cauchy distribution
✤ Even if it looks like a Lorentzian, it might not be a Lorentzian
![Page 44: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/44.jpg)
DEALING WITH GAPS✤ Some solutions are obvious:
✤ Welch method (skip gaps)
✤ Zero padding (or local average)
✤ Other methods are available: Lomb-Scargle
✤ Good for general uneven sampling
✤ Equivalent to linear least-square fit to sin+cos
✤ Statistically robust
![Page 45: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/45.jpg)
LOMB-SCARGLE PERIODOGRAM
✤ hj sampled at tj
✤ where: ensures shift independence
✤ Powerful method: it can go beyond “Nyquist”
![Page 46: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/46.jpg)
BEWARE OF TRENDS!
✤ A trend is a modification to the window
✤ Must be de-trended
✤ Same about possible drop outs
No trend
Trend
![Page 47: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/47.jpg)
CROSS-SPECTRUM✤ Power spectrum: amplitudes of the FFT
✤ We throw away the phases
✤ If we take two time series f(t) & g(t), the phases make more sense
✤ Cross-spectrum:
✤ If f=g, it becomes the power spectrum
✤ What is it useful for?
![Page 48: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/48.jpg)
PHASE/TIME LAGS✤ The phases give us the phase delay between the two
time series
✤ Not easy to interpret, can be linked to physical models
✤ Time lags: phase φ/ν✤ Additional technical
details (not shown)
![Page 49: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/49.jpg)
AUTO/CROSS-CORRELATION
✤ The power spectrum is the FT of the autocorrelation
✤ Autocorrelation is real and even, power spectrum is real and even
✤ The cross spectrum is the FT of the crosscorrelation
✤ Power- an cross-spectrum contain more information (if you can afford them because of statistics)
Corr(g, g) =� ⇥
�⇥g(t + �)g(�)d� �⇥ |G(f)|2
![Page 50: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/50.jpg)
AUTOCORRELATION✤ Uncorrelated noise: ACF is zero everywhere but at τ=0 [variance]
✤ Biased ACF: dividing by N
✤ Unbiased ACF: dividing by N-|m|
Corr(g, g) =� ⇥
�⇥g(t + �)g(�)d� �⇥ |G(f)|2
![Page 51: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/51.jpg)
CROSSCORRELATION
✤ Uncorrelated series: CCF is zero everywhere
✤ Simple shift: peak somewhere
Lag = 20
![Page 52: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/52.jpg)
WHEN DO I HAVE A LAG?
✤ “The CCF peaks at 0, therefore there is no measurable lag”
✤ NO!
✤ CCF is a superposition ofsinusoids of different periods
✤ Any asymmetry implies a lag
![Page 53: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/53.jpg)
COHERENT SIGNALS: BARYCENTRIC CORR.
✤ The Earth moves and rotates, the satellite also moves
✤ This has an effect on the period (doppler modulation)..
✤ .. and on the absolute phase
✤ Times are corrected to the barycenter of the solar system
✤ Standard routines and ephemeris
✤ Not relevant for aperiodic signals
![Page 54: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/54.jpg)
PERIOD FOLDING I: Χ2 TEST
✤ Photon arrival times tj
✤ For trial period produce phases
✤ Put photon in appropriate phase bin
✤ Test vs. constancy (χ2)
✤ If time bins and not times, easy to generalize
✤ Problem: binning and statistics (few photons?)
![Page 55: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/55.jpg)
PERIOD FOLDING II: Z2 TEST✤ Photon arrival times tj
✤ For trial period produce phases
✤ Compute
where n is the desired number of harmonics
✤ Z is distributed as a χ2 with 2n d.o.f.
✤ Good for small number of photons [Rayleigh test]
![Page 56: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/56.jpg)
COMPLICATIONS✤ There can be a significant period derivative
✤ If your pulsar is in a binary system, there is Doppler effect
✤ Easy to lose a pulsation
✤ Power spectrum smeared, folding as well
✤ Must factorize possible orbit in the solution
✤ Many free parameters
![Page 57: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/57.jpg)
MORE ON TIME-FREQUENCYBowhead whale
![Page 58: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/58.jpg)
ALTERNATIVE TECHNIQUES• The Wigner Distribution
• Signal in the past by the signal in the future!
• Problem: only for Gaussian chirps W is everywhere positive
• You can beat the uncertainty principle, but at the cost..
• ... of generatin additional monstruosities
W (t,!) = 12⇡
Rs⇤(t� 1
2⌧)s(t+12⌧)e
�i⌧!d⌧
![Page 59: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/59.jpg)
THE WIGNER DITRIBUTION
Signal: sum of two chirps
![Page 60: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/60.jpg)
THE WIGNER DITRIBUTIONComparison with the spectrogram
![Page 61: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/61.jpg)
COHEN’S KERNELALL time-frequency representations come from:
C(t,!) = 14⇡2
R R Rs⇤(u� 1
2⌧)s(u+ 12⌧)�(✓, ⌧)e
�i✓t�i⌧!+i✓udu d⌧ d✓
Where φ(θ,τ) is the kernel
Changing the kernel you change the representation
The properties of the representation depend on the proprties of the kernel
![Page 62: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/62.jpg)
![Page 63: TIMING METHODS IN X-RAY ASTRONOMYiac.es/winterschool/2015/media/lectures/Belloni_Timing2.pdf · If incident rate r in is very high, one count every t d Detected rate: Formula is even](https://reader033.vdocuments.pub/reader033/viewer/2022042216/5ebf95af8dbc8710b3519326/html5/thumbnails/63.jpg)
A WORD ON WAVELETS• The resolution element in time-frequency cannot be made
smaller than the minimum
• However, there is no reasonwhy it should be of thesame shape
• You can adapt it to whatyou need where youneed it Time
Freq
uenc
y