Autocorrelation

The simplest and most common way to extract information about an optical pulse, other than measuring its power spectrum, is to perform an autocorre- lation measurement, or more accurately a second order autocorrelation. An autocorrelation trace, with prior knowledge or assumption of the pulse shape, allows the temporal width of a pulse to be determined. The associated error can be relatively large (of the order of 10%) but these traces can be taken in real time and require only a relatively simple apparatus as shown in figure 3.5. The autocorrelation technique uses a Michelson-Morley interferometer to split each incident pulse in to two equal halves and then rejoins them after travelling along separate arms. Both arms have nominally equal length, but one stays fixed while the other oscillates. This results in the two halves of the

3. Comparison of existing techniques 49

Fig. 3.5: Schematic of a (2nd order) autocorrelator.

original pulse travelling collinearly after meeting at the beamsplitter, but with a time varying degree of overlap. The degree of overlap determines the peak intensity incident on the detector.

The detected signal will be directly proportional to the time integral of the squared intensity.

S=

Z _∞ −∞

|(E(t) +E(t+τ))2_|2_{dt (3.5)}

Because of this as the pulses move through a cycle of no overlap through to complete overlap and back again the nonlinear signal traces a flat background value (for τ Àτp, where τp is pulse width),

g = 2

Z _∞ −∞

|E(t)2_|2

which rises to a maximum (at τ = 0) of g = 16

Z _∞ −∞

|E(t)2_|2

and back again.

3. Comparison of existing techniques 50 can be calculated exactly, assuming the shape of the pulses is known. The non linear detectors are typically either a photomultiplier tube (PMT) and non linear crystal combination or a two photon absorbing (TPA) material. The PMT, crystal combination offers superior sensitivity, but is relatively bulky, polarisation dependent, more costly and alignment of the crystal is of critical importance. TPA detectors are often chosen because of their ease of use if sensitivity is not a problem. TPA is determined by intensity only and as such polarisation need not be considered.

Whichever type of detector is chosen, the alignment of the overlapping beams is extremely important. The pulse shape is normally assumed from the shape of the spectrum (a Gaussian spectrum corresponds to a Gaussian temporal profile when Fourier transformed, and so this is often the assumed temporal profile). Only if this assumption is true and if the two beams incident on the detector are collinear over the full length of the scanned delay will the calculated pulse width be correct. To aid the alignment of the two beams an ideal ratio of background signal to peak signal should be achieved before measuring the width of the peak. Fortunately the trace can be observed in real time while the beams are being adjusted and this becomes straightforward in practice. The optimal background to peak ratio depends on the response time of detection. From the equations immediately above the optimal peak to background ratio should be 8:1. This is the case for a detection system of sufficiently fast time response. The autocorrelation is called an interferometric autocorrelation because the individual fringes under the pulse envelope can be resolved. Knowing the central wavelength of the pulse allows the spacing of these fringes to provide the time scale of the trace.

If either the detector or connected electronics are too slow then a time averaged, or intensity, autocorrelation results. This has a 3:1 optimal ratio, and exhibits no fine structure. In an intensity autocorrelation the delay scale must be established by scrolling the static delay arm and equating the distance scrolled to the distance the peak of the trace is seen to move.

The interferometric autocorrelation contains phase information, but only qualitatively so. Second order autocorrelations are ambiguous in that they always have symmetrical profiles, which can be realised intuitively by consid-

3. Comparison of existing techniques 51 Relative Intensity 1 3 8 Delay (τ) Delay (τ)

Fig. 3.6: Interferometric (left) and intensity (right) autocorrelations of a Gaussian pulse.

ering that the autocorrelation is constructed by scrolling one pulse through a duplicate pulse. Because the two pulses are identical, it is not possible to state (from the autocorrelation trace alone) which pulse is ahead of the other tem- porally. The intensity measured for a finite degree of overlap could correspond to the trailing edge of pulse0_A0 _{overlapping with the leading edge of pulse}0_B0_, or vice versa. From the practical perspective of measuring pulse width this is immaterial, only a measure of the width of the autocorrelation peak is needed (FWHM is the standard measure).

The FWHM, taken in conjunction with an assumed pulse shape can be used to calculate temporal pulse width. Ideally this value would correspond to transform limited pulses, (i.e. the temporal pulse width given by measuring the frequency spectrum of the pulse, arbitrarily applying a constant spectral phase and Fourier transforming the result). If this occurs then the assumed pulse shape is validated, otherwise little can be concluded other than that the pulse is ”chirped”.