Electronic Acquisition System Optimization

The timing diagram of the 2 dimensional laser scanning is illustrated in Figure 3.11. For the slow axis, the galvanometer was driven by a triangular wave with a frequency of 28 fps typically, which is the frame rate. The driving waveform of the galvanometer was asymmetric so that only one direction of scanning is used for data acquisition. The driven frequency of the galvanometer can be changed to modify the frame rate. For the fast axis, the resonant mirror was driven by a sinusoidal wave with a fixed frequency of 7.9 kHz. To achieve 28 fps frame rate, both directions of the scanning were used for data acquisition but hysteresis would induce a shift in the image and should be handled carefully. 88% of the full amplitude of the mirror was divided into 512 pixels. The pixel number was chosen such that the field is sampled at the Nyquist rate; in other words, the sampling spacing was half of the lateral resolution. (Oversampling at 4096 pixels x 4096 pixels was also allowed but the frame rate would be 8 times slower.) According to simulations, this sinusoidal scanning would result in a longer pixel time at the edges (14 laser pulses) and a shorter pixel time at the center (7 laser pulses). This uneven pixel time gave rise to difficulty of the signal acquisition optimization. Furthermore, if there are distortions in the optical system, the pixel time could be modified to linearize the pixel

displacement across the whole field. To check the linearization, we could simply image a grid slide with this system.

Figure 3.11 The timing diagram of the 2 dimensional laser scanning under a typical setting (512 pixels x 512 pixels, 28 fps). The number of laser pulses per pixel in the fast axis is estimated.

The nonlinear signals excited by the laser pulses were detected by photomultiplier tubes (Hamamatsu R928P), whose typical rise time is few nanoseconds. With sufficient photons, the photomultiplier tubes were operated in the analog mode. (To detect a lower light level, the photomultiplier tubes can be operate in the photon-counting mode to enhance the signal to noise ratio.) The signals from the photomultiplier were pulse-like, spaced temporally at the laser repetition rate and the signal strength was proportional to its peak current, as shown in Figure 3.12. Then, the signal current was amplified by a trans-impedance amplifier and digitized by a data acquisition card (AlazarTech ATS9440).

Figure 3.12 An example of the signal pulses under broadband acquisition (150MHz).

There is a trade off between the gain and the bandwidth of the trans-impedance amplifier. Amplifiers with a large gain would have a narrower bandwidth and vice versa. Depending on the detection bandwidth, the data sampling would be either synchronized to the laser pulses or unsynchronized, discussed as following.

When the bandwidth of the acquisition system is much higher than the repetition rate so that the signal pulses only slightly overlap with each other, the sampling should be triggered by the laser repetition rate with optimal time delay to sense the respective signal peaks. However, when the acquisition is synchronized to the laser pulses, each signal pulse is excited by only one laser shot and therefore the shot noise would be severe. To enhance the signal to noise ratio and avoid data lost, all the values sampled within the same pixel should be averaged by a computer program.

Due to the uneven pixel time, the realtime average and display would be difficult, especially when the pixel number is changeable.

Instead, if the signal pulses are broadened due to the narrow electronic bandwidth so that each sampled value can represent the average of the adjacent signals, the sampling frequency can be different from the laser repetition rate and the pixel average is not necessary. The total experimental setup would be simplified. However, cross talks would occur between adjacent pixels when the signals are too strong or the bandwidth is too narrow. The electronic bandwidth should be tuned to optimize the

image brightness and resolution.

In our system, we chose an amplifier with 500 voltage gain and 50 MHz bandwidth (Hamamatsu C6438-01). We first checked that if laser synchronization is necessary using this amplifier. Described in Figure 3.13, the laser pulses from a photodiode was first shaped into a sinusoidal wave by a 100 MHz amplifier (LeCroy 1855a) and then shifted in time by a digital delay controller with 0.5 ns resolution (Stanford Research Systems DB64). As shown in Figure 3.14, tuning the delay of the trigger could maximize the signal; however, the difference between the peak and the valley was not obvious. Therefore, we believe the synchronization between the laser pulses and the sampling was not necessary with 50 MHz acquisition bandwidth.

Sometimes, we would try to lower down the bandwidth for realtime smoothing by inserting an electronic 1.9 MHz low pass filters (Mini-Circuit BLP-1.9) between the PMT amplifiers and the acquisition card.

Figure 3.13 The synchronized acquisition.

Figure 3.14 The average value of the whole image changes with the delay of the trigger. The optimal delay would be around 9 ns. Longer delay is not necessary since the repetition rate is 96 MHz.

To show the full dynamic range of the detector, the offset and the gain of signal was tuned to match the input range of the analog to digital converter [69]. The minimal gray value was set at the noise level and the maximal gray value corresponded to the strongest signals. Over-saturation should be avoided. Typically, the linear response range of PMTs is about four orders of magnitude so that at least a 14-bit analog to digital converter is necessary.

Due to the photon nature, the measured signal in the same pixel is Poisson distributed. In other words, the pixel value would vary with time and its signal to noise ratio is proportional to the square root of the signal. With strong signals, the pixel value would seem fixed. However, in our case, a high frame rate and a low pixel dwell time resulted in dim images, which suffered a lot from the shot noise. The frames could be averaged in real time with the same frame rate to reduce shot noise.

As shown in Figure 3.15, frame-average enhances the quality of the dim second harmonic light while the bright third harmonic signals do not change much.

(a) (b)

Figure 3.15 A human skin image taken under 28 fps (a) without aveaging and (b) with 10 frame averaging. Green: SHG; Magenta: THG.