Rapid and economical data acquisition in ultrafast frequency-resolved spectroscopy using choppers and a microcontroller

Spectrometers and cameras are used in ultrafast spectroscopy to achieve high resolution in both time and frequency domains. Frequency-resolved signals from the camera pixels cannot be processed by common lock-in amplifiers, which have only a limited number of input channels. Here we demonstrate a rapid and economical method that achieves the function of a lock-in amplifier using mechanical choppers and a programmable microcontroller. We demonstrate the method’s effectiveness by performing a frequencyresolved pump-probe measurement on the dye Nile Blue in solution. © 2016 Optical Society of America OCIS codes: (300.6240) Spectroscopy, coherent transient; (300.6500) Spectroscopy, time-resolved; (300.6530) Spectroscopy, ultrafast; (120.1880) Detection; (120.6200) Spectrometers and spectroscopic instrumentation. References and links 1. R. W. Schoenlein, W. Z. Lin, E. P. Ippen, and J. G. Fujimoto, “Femtosecond hotcarrier energy relaxation in GaAs,” Appl. Phys. Lett. 51(18), 1442–1444 (1987). 2. R. Merlin, “Generating coherent THz phonons with light pulses,” Solid State Commun. 102(2–3), 207–220 (1997). 3. A. M. Weiner and E. P. Ippen, “Femtosecond excited state relaxation of dye molecules in solution,” Chem. Phys. Lett. 114(5–6), 456–460 (1985). 4. M. Cho, “Coherent two-dimensional optical spectroscopy,” Chem. Rev. 108(4), 1331–1418 (2008). 5. D. G. Cahill, W. K. Ford, K. E. Goodson, G. D. Mahan, A. Majudar, H. J. Maris, R. Merlin, and S. R. Phillpot, “Nanoscale thermal transport,” J. Appl. Phys. 93(2), 793–818 (2003). 6. D. G. Cahill, P. V. Braun, G. Chen, D. R. Clarke, S. Fan, K. E. Goodson, P. Keblinski, W. P. King, G. D. Mahan, A. Majumdar, H. J. Maris, S. R. Phillpot, E. Pop, and L. Shi, “Nanoscale thermal transport. II. 2003– 2012,” Appl. Phys. Rev. 1(1), 011305 (2014). 7. A. M. Weiner, Ultrafast Optics (John Wiley and Sons, Inc., 2009). 8. R. Stevens, A. Smith, and P. M. Norris, “Signal analysis and characterization of experimental setup for the transient thermoreflectance technique,” Rev. Sci. Instrum. 77(8), 084901 (2006). 9. J. Zhu, D. Tang, W. Wang, J. Liu, K. W. Holub, and R. Yang, “Ultrafast thermoreflectance techniques for measuring thermal conductivity and interface thermal conductance of thin films,” J. Appl. Phys. 108(9), 094315


Introduction
Ultrafast optical spectroscopy is widely applied in physics and chemistry for time-domain study of energy carrier dynamics and structural information in materials [1][2][3][4], and in thermal engineering for measuring thermal conductivity and interface thermal conductance [5,6].The simplest version of ultrafast spectroscopy involves excitation of the sample by a powerful short pulse called the pump, and tracking the reflection or transmission dynamics of a weaker short pulse called the probe.The probe is delayed with respect to the pump using mechanical translation stages.More sophisticated ultrafast experiments control additional variables such as wavelength and pulse timing, but generally require detecting excitation-induced change in the probe field [7].The pump-probe technique is discussed as an example in this work, although our method is readily applied to more complex spectroscopies.
In a pump-probe experiment, a single-channel photodiode may be used to measure the intensity of the signal field heterodyned with the probe field.In order to remove the effects of environmental background noise, as well as portions of the pump beam scattered into the detector, the pump or both the pump and the probe are usually modulated by mechanical choppers, acoustic-optic or electro-optic modulators.The resulting signal then appears modulated at a frequency which depends on the modulating scheme employed.The modulation signal is provided as the reference input to a lock-in amplifier, which takes the photodiode output and extracts the amplitude and the phase (relative to the reference) of the signal appearing at the reference frequency.This signal processing method is most useful in experiments where the probe is narrowband around a single wavelength and only optical transitions near this wavelength are of interest.One such example is the use of time-domain thermoreflectance to measure thermal properties, which usually involves probing at a single wavelength [8][9][10].
However, when frequency resolution within a broad spectrum is required, as when there are multiple optical transitions of interest, detection by a single-channel photodiode is not ideal.Of course, a monochromator, which is better than optical parametric oscillator (OPO) or amplifier (OPA) [11,12] and common bandpass filters [13] in terms of frequency resolution, can be used to select a narrowband probe at a certain wavelength, and photodiodebased experiments can be repeated at each wavelength.This method is, however, timeconsuming.A better solution is to resolve the whole spectrum of the signal heterodyned with the probe using a spectrometer with a camera.In this strategy, different frequency components in the heterodyned signal are dispersed inside the spectrometer onto different pixels of the camera and frequency-resolved with high resolution.
Unfortunately, common cameras offer no hardware access to signals on individual pixels.Besides, common commercial lock-in amplifiers have only a limited number of input channels (less than 10), so that there is no way to process the signal from every pixel by a lock-in amplifier.Therefore there is a need for an alternative device and accompanying algorithm which together function like a lock-in amplifier.One simple solution is to use shutters to modulate the beams, just like slow choppers except that the user actively controls their opening and closing.The signals with shutters open and closed are collected, and numerically manipulated to extract the material response and remove noise [14,15].This solution suffers from the limited lifetime of mechanical shutters, and especially from low efficiency since shutters take several milliseconds or longer to open or close, during which several pulses are wasted for laser systems with repetition rates at 1 kHz or greater.For improved efficiency, faster methods have been proposed [16,17] which use choppers and a voltage acquisition board, such as a National Instruments (NI) DAQ, to passively read the chopper status.
Here we describe and demonstrate the implementation of a rapid data acquisition scheme for ultrafast frequency-resolved spectroscopy similar to that described by Heisler et al. [17] but with a widely available, economical, programmable and compact microcontroller (Arduino Uno).We discuss the implementation in details and compare results of a frequencyresolved pump-probe experiment on Nile Blue, in which the data is acquired by the microcontroller and a CCD (charge-coupled device) camera versus by a photodiode and a lock-in amplifier.We find that our proposed method can achieve good signal-to-noise ratio with sufficient efficiency, despite that it cannot benefit from the sophisticated electronic noise filters of a typical lock-in amplifier.

Hardware setup for pump-probe measurements
We demonstrate the proposed data acquisition method by performing degenerate pump-probe measurements.The pump and the probe are of the same color and polarization, so that the noise caused by scattered pump light cannot be removed by bandpass filters nor polarizers.A commercial amplified femtosecond laser (Astrella, Coherent, Inc.) provides pulses centered at 800 nm with 1 kHz repetition rate.The 800 nm light pumps a home-built noncollinear optical parametric amplifier (NOPA), generating pulses in the visible range, which are then used for both the pump and the probe.
Figure 1 illustrates the device connection and synchronization scheme.The synchronization and delay generator (SDG) associated with the laser can output transistortransistor logic (TTL) signals with repetition rates of 1 kHz or integer fractions of 1 kHz, synchronized with the laser pulses.We have configured two SDG channels: one provides a 500 Hz signal to drive two choppers, and the other provides the undivided 1 kHz signal to trigger shot-to-shot acquisition by the CCD.One of the choppers uses one half of the frequency of this signal (i.e.250 Hz) as its driving frequency, while the other uses 500 Hz undivided.Each chopper is equipped with an optical sensor that detects the wheel's real-time position and generates a 50% duty-cycle TTL as a feedback for frequency and phase stabilization.We monitor this signal to indicate the laser beam status, i.e. blocked or passed, in real time as it is modulated by the chopper.The CCD is configured as has been described by Augulis et al. [16], and to our knowledge common commercial CCD detectors can be similarly configured.Briefly, the acquisition cycle of the CCD is set as 1 ms so that it can be synchronized with the 1 kHz trigger from the SDG and a number of laser pulses can each be captured without pause during the CCD running time.Thus the two choppers and the CCD are all synchronized with the laser and there is a fixed phase relation between the operations of any pair of these devices.During its running time, the CCD in turn provides a burst of TTL pulses, each synchronized with a single acquisition cycle, i.e. the exposure and the readout of a single frame.As shown in Fig. 1, the TTL signals from the chopper sensors and the TTL burst signal from the CCD are all connected to the digital pins of the microcontroller, an Arduino Uno board in our implementation.The microcontroller is programmable as an independent computer with analog-to-digital conversion modules.Its function here is to read the status of the choppers at the moment when the CCD acquires a frame.This task is accomplished using a hardware interrupt of the processor on the microcontroller.An interrupt is a signal sent by either hardware or software to notify the processor of an event even while it is running another process.This is a basic function of most microcontrollers.The microprocessor can be programmed such that it will immediately exit the current running code upon receiving an interrupt and enter a sub-routine called the interrupt service routine (ISR).The Arduino Uno has two digital pins capable of accepting external hardware interrupts.We wire one of these to the output from the CCD, which provides the TTL burst synchronized with frame acquisition as shown in Fig. 1.The microcontroller is programmed so that a voltage rise at the interrupt pin is interpreted as an interrupt event (alternatively, interrupts could be triggered by a voltage fall or change).Therefore, when the CCD is reading a frame, the rising edge of the synchronized pulse is captured by the interrupt pin and the microcontroller switches to the ISR.In the ISR, voltages are read at pins D0 and D1, which are wired to the output from the chopper feedback sensors.The clock speed of an Arduino Uno is 16 MHz, so that it reacts to interrupts within several microseconds, much faster than the chopping periods (4 and 2 ms), which ensures the status is read accurately.Similarly, the laser pulses (as detected on a photodiode), the TTL pulses from the SDG, and the TTL burst pulses from the CCD are all synchronized with a delay much smaller than the chopping periods.Thus, provided that the chopper is correctly phased and its sensor properly aligned, the reading by the microcontroller reflects whether the laser pulses are blocked or passed in real time.
The chopper feedback TTL signal V f should be low (V f = 0) when a laser pulse is passed and be high (V f = 1) when a laser pulse is blocked (of course this may be reversed, with 0 meaning blocked and 1 meaning passed).This is checked on an oscilloscope by viewing the output of a photodiode placed in the beam path after the chopper, together with the sensor feedback signal, using the latter as the oscilloscope trigger.Figure 2 demonstrates the influence of chopper alignment on the detection of pulse status.The feedback signal is produced by the optical sensor installed on the chopper as shown by dashed red circles in Fig. 1, and marked as a red dot in Fig. 2. A beam going through the green dot is passed when the sensor is exposed with V f = 1, while the beam is blocked when the sensor is blocked with V f = 0.For a beam going through the yellow dot, the situation is similar except that the V f = 0 when the pulses are passed.However, if the beam goes through the blue dot, then reading V f cannot unambiguously detect the pulse status.Care is therefore required when aligning the position of the beam relative to the chopper sensor.The delay or the phase of the chopper relative to the SDG TTL signal, and thus to the laser pulses, can be easily tuned on the chopper controller front panel.Figure 3 shows the well-synchronized readouts of the chopped laser spectrum from the CCD and the pulse status from the microcontroller after configuring the setup as discussed.During examination of the signal timing as shown on the right side of Fig. 2, the time scale of the oscilloscope should be on the order of 1 ms/grid to display multiple periods of the feedback signal on the screen.At such scales, the fast response (~1 ns) from the photodiode may not be well captured because common oscilloscopes will reduce the sampling rate automatically when sampling time range is increased.One trick to overcome this issue is to increase the beam power, saturating the photodiode so that the electronic pulses from the photodiode have much longer duration.Another solution is to adjust the input impedance of the oscilloscope, for instance to 1 MΩ.
In our setup, the pulsed laser has a repetition rate of 1 kHz and the two choppers modulate the pump and the probe at 250 Hz and 500 Hz, respectively.With such a design, every four successive frames acquired by the CCD correspond to a sequence of four intensity values: I 00 for pump on, probe on; I 01 for pump on, probe off; I 10 for pump off, probe on; I 11 for pump off, probe off.The desired signal is then proportional to I 00 -I 01 -I 10 + I 11 .The microcontroller is programmed to digitally read and send the pulse status as 1 or 0 to a computer, which also accepts the frames from the CCD and performs this numerical calculation.For each delay between the pump and the probe, we acquire 4N (N is an integer) successive frames and therefore get N data points, which are then averaged for recording.In practice, since the modulation is perfectly periodic, the microcontroller only needs to read and send the status of the first four frames in any consecutive set.

Pump-probe measurement results
The pump-probe measurements are performed on a solution of Nile Blue dye in ethanol with a concentration of 67 µM contained in a 1 mm-thick cuvette, with a photodiode and lock-in detection and alternatively with a CCD and the proposed data acquisition method.For fair comparison, the experiment setup before the sample is unaltered.The absorption spectrum of the sample and the normalized laser spectrum are shown in Fig. 4 (a) while the normalized autocorrelation trace of the laser pulses is shown in Fig. 4 (b) together with a Gaussian fit, which indicates a pulse duration of 29 fs.The pump fluence is kept at 183 µJ/cm 2 .For lock-in detection, the signal heterodyned with the probe beam propagates through a 40 nm wide bandpass filter centered at 600 nm and is sent into the photodiode (DET210, Thorlabs, Inc.).The pump and the probe are chopped at 357 and 500 Hz respectively with the 7/5 two-frequency chopper wheel.The mixing signal at the sum frequency 857 Hz is used as the lock-in reference.The time constant and sampling rate of the lock-in amplifier are 300 ms and 512 Hz, respectively.Each data point is averaged 80 times, rendering a total time of 8 minutes to acquire 801 data points for 801 delays with step size 6.7 fs, plotted in blue in Fig. 5(b).For CCD detection, the signal heterodyned with the probe is dispersed by a spectrometer onto the CCD (PIXIS100, Princeton Instruments).As described in the previous section, the pump and the probe are modulated by 250 Hz and 500 Hz respectively.Each data point at a specified pump-probe delay is obtained by taking 200 successive frames with the CCD, averaging by 200/4 = 50 and repeating this process by 8 times for further averaging.It takes 104 minutes for acquisition of 801 spectra as shown in Fig. 5(a).The signals at pixels corresponding to the same spectral range (580 to 620 nm) as in the lock-in detection are summed up and shown in Fig. 5(b) for comparison with the lock-in detection.These plots feature photo-induced bleaching via the S 0 -S 1 vibronic transition in Nile Blue.Weak oscillations with a period of roughly 60 fs appear in both measurements, which decay within 0.5 ps after excitation and can be attributed to an impulsively excited ring distortion vibrational mode [18].Because the laser spectrum is narrower than the sample absorption band (Fig. 4 In order to compare the noise levels in the two measurements, 50 successive data points are taken at negative delay after normalization, where there is no sample response.The standard deviation for this noise is 0.0017 for the photodiode and lock-in detection and 0.0019 for the CCD and microcontroller detection.Considering the experimental time cost and noise level, the photodiode and lock-in detection is significantly more efficient if responses at only a few transitions are concerned.This is not surprising since the lock-in amplifier has a sophisticated internal signal-processing apparatus.However, the lower efficiency of the CCD detection is offset by its ability to simultaneously resolve responses from multiple optical transitions.

Conclusion and discussion
We have implemented a rapid data acquisition method for ultrafast frequency-resolved spectroscopy, for which lock-in amplifiers cannot easily be applied.The method is based on a spectrometer with a CCD detector, two choppers and a compact, economical, programmable microcontroller.We demonstrate the performance of this method with pump-probe measurements and compare it with lock-in detection.Although lock-in detection is still preferred when applicable, our method offers commensurate signal-to-noise while measuring all probe frequencies simultaneously, a significant advantage when material responses within a broad spectrum are required with high frequency resolution.Our method is evidently more efficient and robust than shutter-based acquisition, and it is especially advantageous for multidimensional spectroscopy, in which frequency-resolved measurements are repeated for different pulse delays.Furthermore, the functions performed by the microcontroller can be flexibly modified by uploading different programs.Since the microcontroller itself is a programmable computer, it can run independently from the main computer during data acquisition, which helps to release the working load of the CPU and the programing difficulty in correlating the chopper status with data acquisition of the CCD in real time.Compared with other acquisition boards such as NI DAQ, which usually costs hundreds of dollars, the $20 microcontroller is much cheaper and thus is especially useful for groups with multiple ultrafast spectroscopy systems.

Fig. 1 .
Fig. 1.Device connection and synchronization scheme.SDG: synchronization and delay generator; D0(1): digital input pin with index 0(1); INT: digital input pin for hardware interrupt.The dashed red circles indicate the positions of the feedback optical sensors on the choppers.

Fig. 2 .
Fig.2.Influence of beam alignment on the reading of pulse status, blocked or passed.In this example the laser repetition rate is four times the chopping frequency.

Fig. 3 .
Fig. 3. 80 frames of synchronized readouts from the CCD and the microcontroller for one chopped beam.The pulsed laser has a repetition rate of 1 kHz and the chopper is working at 250 Hz.The upper plot shows the frequency-resolved intensity detected by the CCD and the lower shows the corresponding chopper status.The chopper sensor is aligned such that the feedback signal is one when the light is blocked.

Fig. 4 .
Fig. 4. (a) Absorption spectrum of Nile Blue in ethanol (left axis) used in the test and the laser spectrum (right axis); (b) Autocorrelation trace of the laser pulse and the Gaussian fit, which indicates a pulse duration of 29 fs.
(a)), calculating the relative change of transmission ΔT/T in the frequencyresolved results leads to artifacts outside the laser spectral coverage.Therefore in Fig. 5(a) we show the normalized absolute change of the transmitted probe intensity.For the frequencyintegrated curves in Fig. 5(b), we have divided the signal by the incident probe intensity and plot ΔT/T normalized to its maximum.The peak signal in Fig. 5(b) corresponds to ΔT/T = 20%.The two signals in Fig. 5(b) show different responses within 500 fs around the maximum probably because the spectra captured in the two strategies are not perfectly identical.

Fig. 5 .
Fig. 5. (a) Pump-probe measurement results (normalized ΔT) on the solution of Nile Blue dye in ethanol measured by a CCD with the proposed data acquisition method, normalized by the maximum value near 0 ps around 600 nm.(b) Comparison of pump-probe measurement results (normalized ΔT/T) by the two methods: the blue curve indicates the signal measured by a photodiode (PD) with lock-in detection, and is vertically shifted for clarity; the red curve indicates the sum of the frequency-resolved signals from 580 to 620 nm in Fig. 5 (a).