14.

Measuring Ultrashort Laser Pulses I:

Autocorrelation
The dilemma
The goal: measuring the intensity and phase vs. time (or frequency)
Why?
The Spectrometer and Michelson Interferometer
1D Phase Retrieval
Autocorrelation
1D Phase Retrieval
E(t)
Single-shot autocorrelation
E(t–)
The Autocorrelation and Spectrum
Ambiguities
Third-order Autocorrelation
Interferometric Autocorrelation 1
The Dilemma
In order to measure
an event in time,
you need a shorter one.

To study this event, you need a

strobe light pulse that’s shorter.
Photograph taken by Harold Edgerton, MIT

But then, to measure the strobe light pulse,

you need a detector whose response time is even shorter.

And so on…

So, now, how do you measure the shortest event?

2
Ultrashort laser pulses are the shortest
technological events ever created by humans.
It’s routine to generate pulses shorter than 10-13 seconds in duration,
and researchers have generated pulses only a few fs (10-15 s) long.

Such a pulse is to one second as 5 cents is to the US national debt.

Such pulses have many applications in physics, chemistry, biology, and

engineering. You can measure any event—as long as you’ve got a pulse
that’s shorter.

So how do you measure the pulse itself?

You must use the pulse to measure itself.

But that isn’t good enough. It’s only as short as the pulse. It’s not shorter.

Techniques based on using the pulse to measure itself are subtle. 3

Why measure an ultrashort laser pulse?
To determine the temporal resolution of an experiment using it.

To determine whether it can be made even shorter.

To better understand the lasers that Excitation to excited state
emit them and to verify models
Emission
of ultrashort pulse generation.
Excited
state
To better study media: the better
we know the light in and light
out, the better we know the
medium we study with them. Ground state

To use pulses of specific intensity

and phase vs. time to control As a molecule dissociates,
chemical reactions: “Coherent control.” its emission changes color
(i.e., the phase changes),
To understand pulse-shaping efforts revealing much about the
for telecommunications, etc. molecular dynamics, not avail-
able from the mere spectrum,
Because it’s there. or even the intensity vs. time.
4
Studying Media by Measuring the
Intensity and Phase of Light Pulses
Measuring the intensity and phase of the pulses into and out of a
medium tells us as much as possible about the linear and nonlinear
effects in the medium.

With a linear medium, we learn the medium’s

absorption coefficient and refractive index vs. 

Intensity

Phase
E˜ out ( )  E˜ in ( ) exp[  ( )L /2  i k n( )L]

ar
ne
Time (fs)

Li
Intensity

Linear or nonlinear
Phase

medium Exp’t

N
on
lin
Time (fs)

ea
r
With a nonlinear-optical medium, we can
learn about self-phase modulation, for example, Theory

for which the theory is much more complex.

Indeed, theoretical models can be tested.
Eaton, et al., JQE 35, 451 (1999).
5
We must measure an ultrashort laser pulse’s
intensity and phase vs. time or frequency.
A laser pulse has the time-domain electric field:

E(t) = Re { I(t)1/ 2 exp [ i t – i  (t) ] }

Intensity Phase

Equivalently, vs. frequency:

~
E() = Re { S()1/ 2 exp [ i  (–0) ] }

Spectrum Spectral
Phase

Knowledge of the intensity and phase or the spectrum and spectral phase
is sufficient to determine the pulse.
6
The phase determines the pulse’s frequency
(i.e., color) vs. time.
The instantaneous frequency: t ddt

Example: “Linear chirp”

Phase, (t)

time
Frequency, (t)

We’d like to be able to measure,

not only linearly chirped pulses,
but also pulses with arbitrarily complex
time phases and frequencies vs. time.
7
time
Pulse Measurement in the Frequency Domain:
The Spectrometer
The spectrometer measures the spectrum, of course. Wavelength varies
across the camera, and the spectrum can be measured for a single pulse.

Broad-
Entrance
band Slit Collimating
pulse Mirror

“Czerny-Turner”
Grating
arrangement

Focusing
Mirror
Camera or
Linear Detector Array

There are numerous different arrangements for the

8
optics of a spectrometer. This is just one example.
 One-dimensional phase retrieval
It’s more interesting than it appears to ask what information we lack
when we know only the pulse spectrum.

Spectrum


2
S( )  E˜ ( )
Recall: E˜    E t  e  i t
dt and

 ( )  phase[E˜ ( )]

Clearly, what we lack is the spectral phase. Spectral phase

But can we somehow retrieve it?

Mathematically, this problem is called the 1D phase retrieval problem.
Obviously, we cannot retrieve the spectral phase from the mere spectrum.
But what if we have some additional information?
What if we know we have a pulse, which is, say, finite in duration?
There are still infinitely many solutions for the spectral phase.
The 1D Phase Retrieval Problem is unsolvable.

E.J. Akutowicz, Trans. Am. Math. Soc. 83, 179 (1956)

E.J. Akutowicz, Trans. Am. Math. Soc. 84, 234 (1957) 9
Pulse Measurement in the Time Domain:
Power meters
Power meters are devices that emit electrons in response to photons.
Examples: Photo-diodes, Photo-multipliers
power meter
power meter
Another symbol
for a power meter:

Power meters have very slow rise and fall times: ~ 1 nanosecond.

As far as we’re concerned, they have infinitely slow responses.

They measure the time integral of the pulse intensity from – to +:


Vdetector 

E(t)
2
dt

The detector output voltage is proportional to the pulse energy.

10
By themselves, power meters tell us little about a pulse.
Pulse Measurement in the Time Domain:
Varying the pulse delay
Since detectors are essentially infinitely slow, how do we make time-
domain measurements on or using ultrashort laser pulses?

We’ll delay a pulse in time. Input

Mirror pulse E(t)
And how will we do that? Output
pulse E(t–)
By simply moving a mirror! Translation stage

Moving a mirror backward by a distance L yields a delay of:

Do not forget the factor of 2!

  2 L /c Light must travel the extra distance
to the mirror—and back!

Since light travels 300 µm per psec, 300 µm of mirror

displacement yields a delay of 2 ps. Controllable delay
11
steps of less than 1 fs are not too difficult to implement.
We can also vary the delay using
a mirror pair or corner cube.
E(t) Input
Mirror pairs involve two
pulse
reflections and displace Mirrors
the return beam in space: Output
But out-of-plane tilt yields E(t–) pulse
a nonparallel return beam. Translation stage

Corner cubes involve three reflections and also displace the return
beam in space. Even better, they always yield a parallel return beam:

“Hollow corner cubes” avoid propagation through glass. 12

Pulse Measurement in the Time Domain:
The Michelson Interferometer Input
 pulse
VMI ( )
VMI ( ) 
 
E(t)  E(t   ) dt
2

Mirror
E(t)

Slow

Beam- E(t–)


splitter detector
 E(t)  E(t   )  2 Re[E(t)E (t   )] dt
2 2 *
Delay
 Mirror

 
 VMI ( )  2  E (t ) dt  2 Re  E (t ) E *(t   ) dt
2
 

{
The FT of the field
 Pulse energy Field autocorrelation
autocorrelation is
(boring) (maybe interesting, but…) just the spectrum!

Measuring the interferogram is equivalent to measuring the spectrum.

13
Okay, so how do we measure a pulse?
Result: Using only time-independent, linear filters, complete
characterization of a pulse is NOT possible with a slow detector.

Translation: If you don't have a detector or modulator that is fast

compared to the pulse width, you CANNOT measure the pulse
intensity and phase with only linear measurements, such as a
detector, interferometer, or a spectrometer.

V. Wong & I. A. Walmsley, Opt. Lett. 19, 287-289 (1994)

I. A. Walmsley & V. Wong, J. Opt. Soc. Am B, 13, 2453-2463 (1996)

We need a shorter event, and we don’t have one.

But we do have the pulse itself, which is a start.
And we can devise methods for the pulse to gate
itself using optical nonlinearities.

14
Pulse Measurement in the Time Domain:
The Intensity Autocorrelator
Crossing beams in an SHG crystal, varying the delay between them,
and measuring the second-harmonic (SH) pulse energy vs. delay
yields the Intensity Autocorrelation:
Input Aperture eliminates input pulses
pulse and also any SH created by
Mirror the individual input beams.
Beam-splitter
SHG
crystal Slow
E(t)
detector
Vdet ( )  A ( )
(2)
Mirrors E(t–)
Lens ESH (t, )  E(t)E(t   )
Delay ISH (t, )  I(t)I(t   )


The Intensity Autocorrelation: A ( ) 

(2)

 
I(t)I(t   ) dt
15
 Practical Issues in Autocorrelation
Group-velocity mismatch must be negligible, or the measurement
will be distorted. Equivalently, the phase-matching bandwidth must
be sufficient. So very thin crystals (<100 µm!) must be used.
This reduces the efficiency and hence the sensitivity of the device.

Conversion efficiency must be kept low, or distortions due

to “depletion” of input light fields will occur.

The beam overlap in space must be maintained as the delay is scanned.

Minimal amounts of glass must be used in the beam before the crystal
to minimize the GVD introduced into the pulse by the autocorrelator.

It’s easy to introduce systematic error. The only feedback on the 

measurement quality is that it should be maximal at  = 0 and
symmetrical in delay: 

A (2) (  )  A (2)( ) because  I(t)I(t   ) dt   I( t  )I( t) dt

t  t  
16
 Square Pulse and Its Autocorrelation

Pulse Autocorrelation

     
 
FWHM

1; t 2 
1  ;   FWHM
It    A       AFWHM
2 
p A


0;
 t    FWHM
p 2 0;     FWHM
 A


FWHM
p


FWHM
A

t 

 A   p
FWHM FWHM

17
Gaussian Pulse and Its Autocorrelation

Pulse Autocorrelation

  2 ln2t 2    2 ln2  

2

It   exp   FWHM   A    exp   FWHM  

2 
    A  
  p
  
  

 p
FWHM

 A
FWHM

t 

 A  1.41   p
FWHM FWHM

18
 Sech2 Pulse and Its Autocorrelation
Pulse Autocorrelation
A   
2 

1.7627 t  2.7196 
It   sech  FWHM 
2 3 2.7196  
  FWHM  1
t p
coth
 2.7196     FWHM
  A  
sinh 2  FWHM   A
  A 

 p
FWHM

 A
FWHM

t 
  1.54 
FWHM FWHM
A p

Since theoretical models for ideal ultrafast lasers often predict sech2
pulse shapes, people used to (and some still do) simply divide the
autocorrelation width by 1.54 and call it the pulse width. Even when
the autocorrelation is Gaussian… 19
 Sech2 pulse vs Gaussian pulse

the autocorrelation of a
10 fs sech2 pulse the autocorrelation of an
11 fs gaussian pulse
0.5

0
-30 -20 -10 0 10 20 30

20
Lorentzian Pulse and Its Autocorrelation

Pulse Autocorrelation

1 1
It   A   
2 

1  2t    1  2   AFWHM 

FWHM 2 2
p

 p
FWHM

 A
FWHM

t 

 A  2.0   p
FWHM FWHM

21
A Double Pulse and Its Autocorrelation

Pulse Autocorrelation

A 2     A0 2     sep  
It   I 0 (t)  I0 (t   sep )
2 A0 2     A02     sep 

 sep  sep

t 

where: A0   
( 2)

 I0 (t) I0 ( t   ) dt
22
 Multi-shot Autocorrelation and “Wings”
The delay is scanned over many pulses, averaging over any variations
in the pulse shape from pulse to pulse. So results can be misleading.
Imagine a train of pulses, each of which is a double pulse.
Suppose the double-pulse separation varies:

Infinite Train of Pulses Autocorrelation “Coherence

spike”
“Wings”

t 
average larger smaller The locations of the side pulses in
separation separation separation the autocorrelation vary from pulse
to pulse. The result is “wings.”
Wings also result if each pulse in the train has varying structure.
And wings can result if each pulse in the train has the same structure!
In this case, the wings actually yield the pulse width, and the central
spike is called the “coherence spike.” Be careful with such traces.
23
Autocorrelations of more complex intensities

Autocorrelations nearly always have considerably less structure than the

corresponding intensity.

Intensity Autocorrelation

Intensity Autocorrelation
Ambiguous Intensity Ambiguous Autocorrelation

-40 -30 -20 -10 0 10 20 30 40 -80 -60 -40 -20 0 20 40 60 80

Time Delay

An autocorrelation typically corresponds to more than one intensity.

Thus the autocorrelation does not uniquely determine the intensity.

24
Even nice autocorrelations have ambiguities.

These complex intensities have nearly Gaussian

autocorrelations.

Intensity Autocorrelation

Autocorrelation
Intensity Ambig Autocor
Ambiguous Intensity Gaussian

-80 -60 -40 -20 0 20 40 60 80 -150 -100 -50 0 50 100 150

Time Delay

Conclusions drawn from an autocorrelation are unreliable.

25
Retrieving the Intensity from the Intensity Autocorrelation
is also equivalent to the 1D Phase-Retrieval Problem!

A ( )
(2)

 I(t)I(t   ) dt

Applying the Autocorrelation Theorem:

Y { A ( )}  Y {I (t)}
(2) 2

Thus, the autocorrelation yields only the magnitude of the Fourier

Transform of the Intensity. It says nothing about its phase! It’s the
1D Phase-Retrieval Problem again!

We do have additional information: I(t) is always positive. The

positivity constraint reduces the ambiguities dramatically, but still, it
rarely eliminates them all.

The Intensity Autocorrelation is not sufficient to determine the

intensity of the pulse vs. time. 26
Pulse Measurement in Both Domains:
Combining the Spectrum and Autocorrelation
Perhaps the combined information of the autocorrelation and the spectrum
could determine the pulse intensity and phase.

This idea has been called: “Temporal Information Via Intensity (TIVI)”

J. Peatross and A. Rundquist, J. Opt. Soc. Am B 15, 216-222 (1998)

It involves an iterative algorithm to find an intensity consistent with the

autocorrelation. Then it involves another iterative algorithm to find the
temporal and spectral phases consistent with the intensity and spectrum.

Neither step has a unique solution, so this doesn’t work.

27
Ambiguities in TIVI: Pulses with the
Same Autocorrelation and Spectrum
Pulse #1 Pulse #2
Intensity

Phase
Phase

FWHM = 24fs FWHM= 21fs

Intensity
Chung and
Weiner,
Spectra and spectral phases Autocorrelations IEEE JSTQE,
for Pulses #1 and #2 for Pulses #1 and #2 2001.
Spectra
#1

#2

28
These pulses—especially the phases—are very different.
Ambiguities in TIVI: More Pulses with the
Same Autocorrelation and Spectrum
Pulse #3 Pulse #4
Intensity
Chung and
Phase Weiner,
Phase IEEE JSTQE,
FWHM = 37fs FWHM= 28fs 2001.
Intensity

Spectra and spectral phases Autocorrelations

for Pulses #3 and #4 for Pulses #3 and #4
Spectra Despite having
#3 #4 very different
lengths, these
pulses have
the same auto-
correlation and
spectrum!

There’s no way to know all the pulses having a given 29

autocorrelation and spectrum.
 Note the 2
Third-Order Autocorrelation


Third-order nonlinear-optical effects pro-

vide the 3rd-order intensity autocorrelation:

A ( ) 
(3)



I (t)I(t   ) dt
2

k2  0      
Polarization           
2
    E PG
t, E t E t
Gating (PG)  k0  k1  k2  k 2
sig

k1
  (3)


Self-diffrac- k1
 0      
tion (SD)      E SD
t,    E t  2
E t   

k0  2 k1  k2
sig
k
2  (3)


        sig t , 
 PG
Transient   k E
E sig t,    SD
1 0
    TG
k2
Grating (TG)   (3)  k0  k1  k 2  k3 E sig t, 
k  
3

Third-har- k1
   0  3
monic gen-  (3)    E THG
sig t,    E t  2
E t   
eration (THG) k2  k0  2k1  k2

The third-order autocorrelation is not symmetrical, so it yields slightly
more information, but not the full pulse. Third-order effects are weaker,
so it’s less sensitive and is used only for amplified pulses (> 1 µJ). 30
When a shorter reference pulse is available:
The Intensity Cross-Correlation
If a shorter reference pulse is available (it need not be known), then it
can be used to measure the unknown pulse. In this case, we perform
sum-frequency generation, and measure the energy vs. delay.
SFG
E(t) crystal Slow
Unknown pulse detector
Vdet ( )  C( )
Reference Eg(t–)
pulse Lens ESF (t, )  E(t)E g (t   )
Delay ISF (t, )  I(t)Ig (t   )

The Intensity Cross-correlation: C( )   

I(t) I g (t   ) dt

If the reference pulse is much shorter than the unknown

pulse, then the intensity cross-correlation fully determines
31
the unknown pulse intensity.
Pulse Measurement in the Time Domain:
The Interferometric Autocorrelator
What if we use a collinear beam geometry, and allow the autocorrelator
signal light to interfere with the SHG from each individual beam?
Input Developed by
Michelson SHG
Interferometer
pulse Lens crystal Filter Slow J-C Diels
E(t) detector Diels and Rudolph,
Mirror Ultrashort Laser
Pulse Phenomena,
Academic Press,
Beam- E(t–) [E(t)  E(t   )] 2
E(t )  E(t   )
1996.
splitter



Delay
2 2
IA ( )  [E(t)  E(t   )]
(2)
Mirror dt

New Usual
terms  Autocor-


2 relation
IA ( )  E (t)  E (t   )  2E(t )E(t   )
(2) 2 2
dt term


32
Also called the “Fringe-Resolved Autocorrelation”
Interferometric Autocorrelation Math
The measured intensity vs. delay is:


IA(2) ( ) 
 
E (t)  E (t   )  2E(t)E(t   )E
2 2 *2
(t)  E *2 (t   )  2E * (t)E * (t   ) dt

Multiplying this out:



 E (t) 2
IA ( )   E (t) E (t   )  2E (t) E (t) E (t   ) 
(2) 2 2 *2 2 * *

2

E (t   ) E (t)  E (t   )  2E 2 (t   )E * (t)E * (t   ) 
2 *2 2

2E(t)E(t   )E *2 (t)  2E(t)E(t   )E *2 (t   )  4 E(t) E(t   )

2 2
dt



 
I 2
(t)  E (t)E (t   )  2 I(t)E(t)E (t   ) 
2 *2

E (t   ) E (t)  I (t   )  2 I(t   )E (t)E(t   ) 

2 *2 2
*

2 I(t )E(t   )E (t)  2I(t   )E(t)E (t   )  4 I(t)I(t   ) dt

* *

where I(t)  E(t)

2
33
The Interferometric Autocorrelation is the
sum of four different quantities.




I (t)  I (t   ) dt
2 2
Constant (uninteresting)

 4

I(t)I(t   ) dt Intensity autocorrelation


Sum-of-intensities-weighted 
 2 I(t)  I(t   )E(t)E (t   ) dt  c.c
*
“interferogram” of E(t) 
 (oscillates at  in delay)



Interferogram of the second harmonic;
 E (t)E (t   ) dt  c.c.
2 2*
equivalent to the spectrum of the SH 
 (oscillates at 2 in delay)
The interferometric autocorrelation simply combines several measures
of the pulse into one (admittedly complex) trace. Conveniently, however,
they occur with different oscillation frequencies: 0, , and 2. 34
Interferometric Autocorrelation and Stabilization
To resolve the  and 2 fringes, which are spaced by only  and /2,
we must actively stabilize the apparatus to cancel out vibrations, which 
would otherwise perturb the delay by many .

Interferometric Autocorrelation Traces for a Flat-phase Gaussian pulse:

With stabilization Without stabilization

Pulse
C. Rulliere,
length
Femtosecond
Laser Pulses,
Springer,
1998.

Fortunately, it’s not always necessary to resolve the fringes.

35
Interferometric Autocorrelation: Examples
The extent of the fringes (at  and ) indicates the approximate width of
the interferogram, which is the coherence time. If it’s the same as the 
width of the the low-frequency component, which is the intensity 
autocorrelation, then the pulse is near-Fourier-transform limited. 

Unchirped pulse (short) These

Chirped pulse (long)
pulses
have
~ Coherence identical ~ Coherence
time spectra, time
and hence
~ Pulse identical ~ Pulse
length coherence length
times.

Solid black lines have been added. C. Rulliere,

Femtosecond
They trace the intensity autocorrelation Laser Pulses,
component (for reference). Springer,
1998.
The interferometric autocorrelation nicely reveals the approximate pulse
length and coherence time, and, in particular, their relative values. 36
Interferometric Autocorrelation:
Practical Details
A good check on the interferometric autocorrelation is that it should
be symmetrical, and the peak-to-background ratio should be 8.

This device is difficult to align; there are five very sensitive degrees of
freedom in aligning two collinear pulses.

Dispersion in each arm must be the same, so it is necessary to

insert a compensator plate in one arm.

The typical ultrashort pulse is still many wavelengths long.

So many fringes must typically be measured: data sets are large,
and scans are slow.

Like the intensity autocorrelation, it must be curve-fit to an assumed

pulse shape.

37
Does the interferometric autocorrelation yield
the pulse intensity and phase?
No. The claim has been made that the Interferometric Autocorrelation,
combined with the pulse interferogram (i.e., the spectrum), could do so
(except for the direction of time).

Naganuma, IEEE J. Quant. Electron. 25, 1225-1233 (1989).

But the required iterative algorithm rarely converges.

The fact is that the interferometric autocorrelation yields little more

information than the autocorrelation and spectrum.

We shouldn’t expect it to yield the full pulse intensity and phase. Indeed,
very different pulses have very similar interferometric autocorrelations.

38
Pulses with Very Similar Interferometric Autocorrelations
Without trying to find ambiguities, we can just try Pulses #1 and #2:
Pulse #1 Pulse #2
Intensity

Phase
Phase

FWHM = 24fs FWHM= 21fs

Intensity

Interferometric Autocorrelations for Pulses #1 and #2

#1 and #2
Despite the very
different pulses, Chung and
Weiner,
these traces are IEEE JSTQE,
nearly identical! 2001.
Difference:

39
Pulses with Very Similar Interferometric Autocorrelations
It’s even harder to distinguish the traces when the pulses are shorter,
and there are fewer fringes. Consider Pulses #1 and #2, but 1/5 as long:
Pulse #1 Pulse #2
Intensity

Phase
Phase

FWHM=4.8fs FWHM=4.2fs
Intensity
-20 -10 0 10 20 -20 -10 0 10 20

Interferometric Autocorrelations for Shorter Pulses #1 and #2

#1 and #2
In practice, it would be Chung and
virtually impossible to Weiner,
IEEE JSTQE,
distinguish these.
2001.

Difference:

40
More Pulses with Similar Interferometric Autocorrelations
Without trying to find ambiguities, we can try Pulses #3 and #4:
Pulse #3 Pulse #4
Intensity

Phase
Phase
FWHM = 37fs FWHM= 28fs
Intensity

Interferometric Autocorrelations for Pulses #3 and #4

#3 and #4
Chung and
Despite very different pulse
Weiner,
lengths, these pulses have IEEE JSTQE,
nearly identical IAs. 2001.

Difference:

41
More Pulses with Similar Interferometric Autocorrelations
Shortening Pulses #3 and #4 also yields very similar IA traces:
Pulse #3 Pulse #4
Intensity

Phase
Phase
FWHM=7.4fs FWHM=5.6fs
Intensity
-40 -20 0 20 40 -40 -20 0 20 40

Interferometric Autocorrelations for Shorter Pulses #3 and #4

Shortened #3 and #4
pulse (1/5
as long) Chung and
Weiner,
IEEE JSTQE,
2001.
Difference:

42
It is dangerous to derive a pulse length from the IA.