An´alisis de correlaci´on

−40 −20 0 20 40 Power (a.u.) −50 0 50 Time (ps) (a) Chirp (GHz) −50 −25 0 25 50 Delay (ps) (b) AC Power (a.u.)

Figure 6.6: (a) Retrieved powers (solid markers, left ordinates) and chirps (empty markers, right ordinates) of the probe pulses. (b) Overlay of autocorrelation trace from a commercial device (solid line) and numerically generated autocorrelations of the retrieved probe pulses. The probe pulses in the figures were mixed with signal pulses compressed through 95 (green diamonds), 110 (red squares), and 125 m (blue circles) long DCFs.

6.2

Results and Discussions

Noise suppression of the measured spectrograms was achieved by background subtrac- tion and low-pass filtering. They were then interpolated onto a Fourier grid with a temporal span equal to, or less than the pulse period before applying the retrieval al- gorithm described in Section 2.5. The top row of Fig. 6.4 shows a series of interpolated spectrograms for signal pulses compressed with the three different DCFs lengths. Al- though we did not impose a spectral constraint in the retrieval algorithm, no nontrivial ambiguities were observed. With a 128_×128 spectrogram, the algorithm converged rapidly after 50-150 iterations to a FROG retrieval error [Eq. (2.24)] of less than 0.005 in all cases indicating good quality retrievals. Figure 6.5 shows the typical FROG re- trieval error as a function of the number of iterations. The bottom row of Fig. 6.4 shows the retrieved spectrograms.

Figure 6.6(a) shows the retrieved intensities and chirps of the probe pulses. We found a reasonably consistent agreement between the retrieved probe pulses as we used the same probe pulse in all of our measurements. Slight discrepancies in the temporal shape among the retrieved pulses can be attributed to the sampling error induced by our particular motorised optical delay line, whereas the discrepancies in the chirp outside the main pulse arise from the retrieval uncertainty in the low intensities region. Addi- tional comparison between the numerical autocorrelation of the retrieved probe pulses 6.2 Results and Discussions

Power (a.u.) 0 0.3 −0.3 Spectral Powe r (a. u.) 1545 1546 1547 1548 Wavelength (nm) (c) −0.5 0 0.5 −20 −10 0 10 20 Time (ps) Chirp (THz ) −20 −10 0 10 20 Time (ps) −0.3 0 0.3 −20 −10 0 10 20 Time (ps) 1545 1546 1547 1548 Wavelength (nm) (b) 1545 1546 1547 1548 Wavelength (nm) (a)

Figure 6.7: Top: Retrieved powers (red circles, left ordinates) and chirps (red squares, right ordinates) of the signal pulses compared with independent measurements made with a commercial SHG-FROG pulse analyser (blue lines). Bottom: Retrieved (red circles) and measured (blue lines) signal pulse spectra. The signal pulses were compressed with a) 95, b) 110, and c) 125 m long DCFs.

and their autocorrelation traces measured with a commercial autocorrelator, shown in Fig. 6.6(b), gave excellent agreement. In all cases, the width of the probe pulse is 25.5 ps. Finally, we note that the red-shift chirp (the frequency decreases with time across the pulse) observed in the retrieved probe pulse is known to be imposed by the EAM.5

The top row of Fig. 6.7 shows the retrieved powers and chirps of the signal pulses compressed with three different DCFs lengths. We compared the retrieved signal pulses with independent characterisations using a commercial SHG-FROG pulse analyser (”Southern Photonics”). Characterisations using the Southern Photonics device were done before the pulses were combined via a 50:50 coupler. The excellent agreement be- tween them confirms the high reliability of our measurements in all cases. The measured width for the signal pulses compressed with 95, 110, and 125 m of DCFs are 7.5, 6.0, and 4.5 ps, respectively. Gain-switching in DFB lasers is known to cause a red-shifted chirp on the output pulse due to refractive index changes during optical pulse propa- gation,6 and remains of this can be seen as the signal pulse was compressed with 95 m of DCF, suggesting the possibility of further compression using a longer DCF. Indeed, the amount of this red-shift chirp decreased when the signal pulse was compressed with 110 and 125 m long DCFs. The satellite pulses are evidence of residual nonlinear chirp

1542 1544 1546 1548 1550 1552 Wavelength (nm)

Spectral Power (a.u.)

Figure 6.8: Mode-locked fibre laser (Pritel) spectrum.

on the edges of the GS-DFB pulse.6

Comparisons between the retrieved spectra of the signal pulses and direct measure- ments of the spectral envelopes taken from the OSA [bottom row of Fig. 6.7] yield good agreement. The fringes in the measured spectra are associated with the 10 GHz repetition rate of the pulse train. Since the retrieved spectra correspond to a single pulse, such fringes do not exist. Note also that the spectral shape is different for each compression fiber length due to the fact that we adjusted the GS-DFB laser drive and injection seeding conditions to minimise the pulse duration in each instance.

For the results in this experiment, the peak power of the coupled probe pulse is evaluated to be _∼14 mW, whereas the peak powers of the coupled signal pulses com- pressed with 95, 110, and 125 m of DCF are ∼54, 67, and 90 mW, respectively. These values correspond to a device sensitivity of _∼60 mW2. Reducing the input powers by a factor of 3 would still yield sufficient output power for a reasonable measurement, leading to an estimated device sensitivity of better than∼6.7 mW2. It is worth noting that characterisations of both signal and probe pulses after the combining 50:50 coupler using the commercial SHG-FROG device were not possible due to the limited device sensitivity. In addition, the limited resolution of the commercial SHG-FROG device spectrometer made it difficult to obtain a reliable spectrogram.