We have developed a simple detection scheme that uses an 8-bit CMOS camera and spans over 60-dB dynamic range. By use of noise reduction techniques, the 8-bit CMOS camera yields a 40-dB dynamic-range signal, which is further increased by 20 dB by making a replica of the signal beam on another part of the detector chip. We have experimentally validated this scheme in a scanning and a single-shot autocorrelator.
© 2007 Optical Society of America
The development of modern high-energy petawatt laser systems around the world [1, 2] is made possible by applying the chirped-pulse-amplification (CPA) technique  to systems with a large number of amplification stages. After the last stage of amplification, the stretched pulse is usually compressed to sub-picosecond durations before being sent to the experimental area. The adjustment of a compressor with meter-size optics requires an efficient and versatile temporal diagnostic that should be able to work in single-shot mode and at low energy during the alignment of the compressor. Indeed, two temporal parameters are important during the compressor adjustment: the pulse duration which is directly related to the achievable on-target peak intensity and the pulse pedestal. In particular the pulse pedestal has received a growing interest because temporal contrast [4, 5] is crucial for most applications of high-peak power laser pulses.
Many devices have been developed to allow the measurement of pulse durations either directly  or indirectly through a retrieval algorithm  in scanning as well as in single-shot modes [8–10]. Since pulse contrast was of lesser importance and compressor adjustment easy, the dynamic range performance of such devices was never optimized, leaving the high-dynamic- range measurement to the tedious 3-ω scanning cross-correlator. However, investigating the temporal behavior of pulses amplified by new techniques such as OPCPA  as well as the complex alignment of compressors with meter size gratings require the development of a device with a high-dynamic range and the capability of single-shot operation.
A standard non-collinear autocorrelator offers a dynamic range which spans over several 10’s of dB [12–15], theoretically limited by background photon noise from self-doubled scattered light. Experimentally, this limit is not easily met by lack of a simple detector scheme matching that dynamic range. For the detection scheme, the main challenge resides in achieving analog-to-digital conversion (ADC) with the required bit depth.
In this paper, we experimentally demonstrate a simple detection scheme that can be applied to a cross-correlator setup and improves its dynamic range by several orders of magnitude. To achieve the required bit depth, we use a detector array, as a massively parallel measurement device. Detector arrays, whether Charged Coupled Device (CCD) or Complementary Metal Oxide Semiconductor (CMOS), are now available in large array sizes that exceed 1k × 1k pixels. When collectively used for measuring a single value, an 8-bit 1k × 1k detector array offers virtually 28 bits of detection. However, such a measurement is only beneficial if the signal-to-noise ratio of the unit pixel can be improved through the collective use. Therefore, the first part of the paper addresses the increase of the signal-to-noise ratio, which is done along two axes: the first method increases the signal by modulation of the signal intensity such that the detector scans the signal; the second one uses averaging to reduce the noise. Since single-shot operation is required, these operations need to be performed in parallel on the same detector array. The second part of the paper reports on the experimental implementation of the scheme, as the detection for both scanning and single-shot autocorrelators, where one reaches 57-dB and 60-dB dynamic range respectively, matching the theoretical dynamic-range limit of such devices.
2. Method and predictions
The signal modulation is implemented by use of an uncoated wedge with a reflectivity of 0.18 % at 527 nm, in order to generate a replica of the beam with a reduced intensity on the detector array. The wedge creates a large number of beam replicas, all separated in intensity by the same factor. When two signals are analyzed simultaneously, this gives an additional dynamic range of 27 dB; when three signals, the increase in dynamic range is 54 dB. However, the blooming effect that appears on the CCD signal, when one area is highly saturated, is a technical limitation that makes CCD impractical. The CMOS detector is less sensitive to blooming making it more appropriate for the detection scheme. For experimental reasons, we only used a single replica and therefore our signal gain is limited to 27 dB.
However, it has been reported that CMOS suffers more from noise than scientific CCD cameras , and careful noise reduction has to be performed. The noise reduction is based on the averaging technique, but this can only be achieved efficiently when one understands the noise contribution to the signal. The noise can be attributed to two sources: a local contribution σpix and a contribution σcorr correlated across the detector array:
The correlated noise, due to fluctuations of external factors, does not benefit from averaging out. The relative weight of the two terms σpix and σcorr can be analyzed from the graphs in Fig. 1 The graph (a) shows the average pixel value for two different 250 × 250 pixel areas on the CMOS detector chip (Basler A622f camera) recorded simultaneously as a function of the frame number. The second area plot has been offset to increase the visibility of the two plots. The correlated noise is responsible for the correlated evolution of the signals, while the pixel-to-pixel noise characterizes the difference. At first sight, it is clear that the correlated noise dominates the signals. Fig. 1(b) shows the standard deviation of the averaged signal counts, for 1000 frame series, as a function of the pixel number. The log-log plot shows the theoretical limit that is expected from a purely random pixel-to-pixel noise, following the “square root” law. Two types of noise are plotted: first the standard deviation for a single averaging area, and second the standard deviation for the difference between two areas within the same frame (labeled “referenced area” on the plot). Both plots follow the “square root” law for small areas below 10. The benefit from the averaging techniques is clear as the noise reduction is then 10 dB. However, the referenced area method is more efficient for square areas larger than 10 pixels. While a steady state at 12 dB noise reduction is reached for the single-area method, the noise is further decreased when the signal area is referenced to another non-illuminated area of the same frame. For the 250 × 250 pixel square area used later in the paper, the noise is reduced by 18 dB, only 3 dB above the theoretical limit.
As a conclusion, averaging does not help beyond 10 × 10 pixels areas, unless the correlated noise is removed from the frame by utilizing a reference area within the chip. Practically, this corresponds to removing a background term calculated from an unilluminated area (the reference area), scaled by a calibration factor calculated from a dark calibration frame. In such a case, the processed signal Sp will be:
where S, S0, Sr and Sr0 are the signal, the signal of the dark calibration frame, the reference signal and the reference signal of the dark calibration frame, respectively.
This noise reduction, provided it yields a noise at the 0.017 count level, is greatly improving the dynamic range of the CMOS detector. By assuming that the signal fully fills the 255 counts of the detector, the signal-to-noise ratio equals 42 dB. The addition of the replica technique as described above will extend the dynamic range by 27 dB, adding up to a total theoretical dynamic range of 69 dB, if a 250 × 250 pixel detection area is used.
3. Experimental setup and results
Fig. 2 shows a schematic of the experimental setup implemented for the scanning and the single-shot autocorrelators. The details of each autocorrelator scheme are out of the scope of this paper and can be found elsewhere [8–10,17]. In both cases, the detector is an 8-bit CMOS camera - Basler A622f - and the signal is magnified to cover a large area of the chip.
In addition the 1° uncoated wedge is located into the setup, creating a beam replica shifted on the camera [Fig. 3(a)]. Fig. 3(b) shows an example of the resulting camera image with the three different areas used for the detection. In the case of the single-shot autocorrelator, a cylindrical lens, located in front of the setup, is used to control the size of the autocorrelation trace in the vertical direction and therefore prevent cross-talk between the main and replica signals on the CMOS chip.
The detection scheme of the scanning autocorrelator has been validated by measuring the autocorrelation signal from a 150 fs, 300 mW, 76 MHz Ti:sapphire oscillator running at 1053 nm. Fig. 4 shows the results in three different signal retrieval methods. In all cases, the signal is averaged over 10 frames. In the first case, the signal is determined by recording the value of a single pixel without image replica. This autocorrelation trace shows a dynamic range of 27 dB, which is mostly due to the average over 10 frames. The second trace shows the result when the image averaging is performed, but the replicated signal not used. In this case, the dynamic range is increased to 37 dB. The third trace shows the measurement with the replica beam used and yields a dynamic range equal to 52 dB. In this case, the limiting factor comes from the SHG light scattered in to the camera. If one considers the detection noise floor, when the input beam is blocked, the value is 57 dB as shown in Fig. 4.
The prediction stated that the detection background would be 69 dB. Experimentally, we measure a noise background of 37-dB without the replica and 57-dB with the replica. This shows that the noise reduction done by Eq. (2) does not fully remove the noise from the signal and thus explains for the loss of 5 dB. The second experimental limitation is the available power from the laser that does not allow the replica to fully use the 8-bit dynamic range of the camera. Instead, only 20-dB are used, at best.
Similar results were obtained with the single-shot autocorrelator. The light pulse under study originated from the same Ti:sapphire oscillator running at 1053 nm. However, the pulse energy was too low for a direct measurement; and the pulse was thus amplified in a chirped-pulse amplifier before the measurement. After stretching the pulse to 3 ns, its amplification to 5 mJ and the recompression to ~ 500 fs, only ~ 125 μJ were required for the measurement.
Blurring is not totally absent in the CMOS detector. In particular, the area located in a vertical line of the saturated area exhibits a measurement artifact. For this reason, the replica peak center is shifted horizontally to the side of the chip, as shown in Fig. 3(b). Calibration is easily done in two steps: first for background removal, second for relative peak position assessment. The signal being mapped along one direction, the background referencing is done pixel by pixel rather than globally. Although more CPU intensive, this seems to be more efficient in the background removal as it can be seen later. The raw signal is obtained by binding the pixels along the vertical direction. The mapping of time delays along one direction of the chip means that the number of pixels used for the signal measurement is not as large as in the case of the scanning autocorrelator. To mitigate this effect, a temporal convolution of the raw signal is operated to reduce the pixel noise. We limit the correlation window to 100 fs to avoid signal broadening. Fig. 5 shows a typical measurement plot. The autocorrelation peak was offset in order to offer a 20-ps temporal window. After a sharp 30-dB decrease of the intensity in 5 ps, the signal exhibits a slow decrease of 10 dB in the following 15 ps. This behavior was not expected and is attributed to a large spectral clipping in the pulse stretcher. We checked that this is not a measurement artifact by blocking one and then the other arm of the autocorrelator to look for self SHG from each arm, which was absent. The time delays beyond 20 ps are mapped outside the SHG crystal area and show the detection background noise, around 60 dB.
The average number of pixel used for a single-point measurement is 150, which is 400 times less than used for the prediction. From the plot on Fig. 1(b), a 9-dB reduction can be expected. So the prediction for the single-shot autocorrelator is a 60-dB dynamic range, which is the experimental value. For this setup, the amplified pulse energy allows to make full use of the total dynamic range of the detector. So, by setting the maximum count value for the replica beam above 200 counts, the loss from the maximum range is lower than 1 dB. In addition, background removal seems more efficient than in the scanning autocorrelator so the background noise of the signal without replica is equivalent, although a smaller number of pixels are used.
In this paper, we have shown that by using the collective response of the individual pixels of a CMOS detector, one can greatly improve the dynamic range of a standard autocorrelator setup. We have demonstrated that a standard off-the-shelf low-cost 8-bit detector yields up to 60-dB-dynamic-range temporal measurement of short pulses. It is not clear that the use of a detector with a higher bit depth ADC would improve our results; we believe instead that the careful selection of a low noise detector is the solution to improve the result further. In our case, the maximum signal-to-noise ratio of the camera equals 24 dB but other devices can probably exceed this, improving the available dynamic range of this scheme. Although the 60-dB dynamic range does not totally reach the level required for characterizing high-contrast pulses, we have found the autocorrelator an easy-to-use tool for the alignment and early diagnostic of our system.
References and links
1. M. D. Perry, et al., “Petawatt laser pulses,” Opt. Lett. 24, 160–162 (1999). [CrossRef]
2. J. D. Zuegel, et al., “Laser challenges for fast ignition,” Fusion Sci. Technol. 49, 453–482 (2006).
3. D. Strickland and G. Mourou, “Compression of amplified chirped optical pulses,” Opt. Commun. 56, 219–221 (1985). [CrossRef]
4. D. Umstadter, “Review of physics and applications of relativistic plasmas driven by ultra-intense lasers,” Phys. Plasmas 8, 1774–1785 (2001). [CrossRef]
5. P. A. Norreys et al., “PW lasers: matter in extreme laser fields,” Plasma Phys. Controlled Fusion 46, 13–21 (2004) [CrossRef]
6. C. Iaconis and I. A. Walmsley, “Spectral phase interferometry for direct electric-field reconstruction of ultra-short optical pulses,” Opt. Lett. 23, 792–794 (1998). [CrossRef]
7. D. J. Kane and R. Trebino, “Single-shot measurement of the intensity and phase of an arbitrary ultra-short pulse by using frequency-resolved optical gating,” Opt. Lett. 18, 823–825 (1993). [CrossRef]
9. A. Brun et al., “Single-shot characterization of ultra-short light pulses,” J. Phys. D: Appl. Phys. 24, 1225–1233 (1991). [CrossRef]
10. M. Raghuramaiah et al., “A second-order autocorrelator for single-shot measurement of femtosecond laser pulse durations,” Sadhana 26, 603–611 (2001). [CrossRef]
11. N. Forget et al., “Pump-noise transfer in optical parametric chirped-pulse amplification,” Opt. Lett. 30, 2921–2923 (2005). [CrossRef]
12. O. Konoplev et al., “Ultrahigh dynamic range measurement of high-contrast pulse using second-order autocorrelator,” LLE Review 75, 159–170 (1998), http://www.lle.rochester.edu/03_publications/03_01_review/pastreviews/lle-review-75.html
14. P. F. Curley, et al., “High dynamic range autocorrelation studies of a femtosecond Ti:sapphire oscillator and its relevance to the optimisation of chirped pulse amplification systems,” Opt. Commun. 120, 71–77 (1995). [CrossRef]
16. A. Hoffman, et al., “CMOS Detector Technology,” Exp. Astron. 19, 111–134 (2005) [CrossRef]
17. L. Sarger and J. Oberle, How to measure the characteristics of laser pulses in Femtosecond Laser Pulses, C. Rulliere, ed., (Springer-Verlag, New York,2004).