theoretical background
4.2 Instrument design
From the three Raman micro-spectroscopic instruments described in this thesis only the wide-field Fourier transform instrument was designed and built by the author of this thesis. Therefore, only the steps in the design and implementation of this instrument are included.
A photograph of the interferometer is shown in Figure 4.5.
TTL box
R2
R1 CCD
CCD DM
Pd L1
L2
Microscope
Figure 4.5: Interferometer. Abbreviations: Dichroic mirror (DM), lenses(L1, L2), retroreflectors (R1, R2), photodiode (Pd), triggering box (TTL box), and charge coupled device (CCD).
In the following sections a discussion of the optimal components for the imple-mentation of the instrument is presented.
4.2.1 Optical components
A Nd:YAG visible laser with <5 mW and 532 nm excitation wavelength (JPM-X-3, Laser Vision) was used. For the initial work including alignment and testing of the operational principle of the system a visible laser was chosen, as the Raman signal is higher for smaller wavelengths and therefore easier to detect. Using a visible
laser implies that in the interferometer the Jacquinot advantage was present but not the multiplex advantage. In future prototypes designed for skin imaging NIR lasers might be preferred.
Dichroic beamsplitters are non-absorbing optical filters which reflect light from a specific spectral region while allowing optimal transmission at different wave-lengths. 45 degree oriented dichroic beamsplitters reflect light perpendicularly to the incident light. In Raman spectroscopic systems dichroic beamsplitters are used to reject the strong Rayleigh component of the scattered light. Although an ideal dichroic beamsplitter will completely separate the Raman and Rayleigh components in practice a small amount of elastically scattered light is transmitted along with the Raman signal. A 45 degree ultrasteep dichroic beamsplitter with 538.9-824.8 nm passband and < 186 cm−1 transition width (RazorEdge, Semrok) was chosen for the instrument.
In addition, a 25 mm N-BK7 cube beamsplitter (Techspec, Edmund Optics) op-timised for the spectral range of 430-670 nm was used. The coating guarantees reflection and transmission of 50 ± 5%. While plate beamsplitters require the in-troduction of a compensation plate of the same material as the beamsplitter to ensure symmetry between both arms of the interferometer, cube beamsplitters do not.
Broadband hollow retroreflectors (UBBR2.5-1I, Newport, Irvine, USA) of paralle-lism 1 arc sec, 63.5 mm diameter aperture, and total reflectivity R > 73% for the spectral range of 400-650 nm-wavelength were used. In corner cube retroreflectors the reflected beam is parallel to the incident beam regardless of the angle of inci-dence. The retroreflectors were optimised for the NIR (in NIR, R > 94%, coated for 650 to 12000 nm operation). Due to its high cost this piece of equipment was optimised for the final NIR instrument although visible retroreflectors would have been desirable for the first prototype described in this thesis.
4.2.2 Step-motor: rapid-scanning versus step-scanning
Step scanning allows longer exposure times for recording each interferogram as the movable mirror can be stopped and left at each position as long as needed.
This is a key aspect when dealing with weak Raman signals. For skin tissues Raman spectral acquisition times might be as long as 1 s. However, introduction of accelerations that allow the mirror to move and stop intermittently add mechanical and electronic background noise to the interferogram. Rapid scanning reduces this noise as well as lowers the acquisition time, but faster mirror velocities imply
shorter exposure times to collect the Raman signal. In addition, a fast driven mirror requires a detector with faster frame rates normally of higher cost.
4.2.3 Laser reference signal detection and TTL converter box
1. Internal calibration: reference laser.
Both Raman and 532 nm-wavelength laser signals pass through the interfe-rometer. The laser component allowed internal calibration of the spectral axis. When a source of monochromatic wavelength is used and the mirror is moved with constant velocity v, the frequency of the interference signal fint
can be expressed as
fint = 2v
λ (4.1)
As the laser wavelength was 532 nm and the interval of mirror velocities was 5000 (nm)s−1-1 (mm)s−1, fint was estimated to belong to the range of frequencies 18.7-3.8 kHz. If v is larger than 1(mm)s−1, the process is called rapid-scan interferometry [168]. Thus, the working regime used in this thesis is below rapid-scan interferometry.
As lasers are quasi-monochromatic sources the shape of the laser interfe-rogram recorded was sinusoidal. The period of the signal depends on the step-motor velocity v and goes from 0.3 ms to 53.5 ms. A threshold voltage can be used to convert this signal into a TTL signal or square wave. If the detected voltage is larger than a certain value close to the maximum of the signal, a 5 V voltage will be generated. Otherwise, the output voltage of 0 V will be produced. Each time that a maximum of signal (5V) was detected by the CCD, a frame was recorded by the device. Time retardation between consecutive frames was the time delay between two maxima in the TTL si-gnal. The period of the TTL signal was the same as that of the original sine signal. For longer periods a slower detector could be used in terms of frames per second. However longer periods imply slower step-motor velocities and an increase in the acquisition time.
2. TTL conversion circuit
The TTL conversion circuit was built by the electronics workshop of The School of Physics and Astronomy of The University of Nottingham. A de-tailed description is included in Appendix II.
4.2.4 Detectors
The interferometer output beam carries both the Raman contribution and a remaining laser signal. As the Raman signal intensity is much smaller than the laser intensity both components need to be separated for the Raman beam to be detectable. A dichroic beamsplitter was introduced in the op-tical path (LPD01-532RS-25×36×2.0, Semrok) at the interferometer exit.
The Raman signal passed through, while the laser component was reflected towards the photodetector. Both signals were focused on the focal plane of their respective detectors by focal-length optimised convergent lenses. A Si PIN photodiode with a photosensitivity of 0.4 AW−1 for the desired wave-length, an active area diameter of 0.4 mm, and NEP of 1.5 × 10−15WHz−12 was used to detect the laser signal reflected by the dichroic beamsplitter.
A cooled CCD with a maximum of 2-3 frames per second (fps) was also used. The CCD had a quantum efficiency of ≈ 45% at 532 nm, and a dark current of 0.002 electrons per pixel per second at maximum cooling. In this experiment a high responsivity and low dark current were required. The calculation of the three main sources of noise (thermal, shot, and readout) are included.
The register well depth for the CCD is 106 electrons per pixel. Assuming an electron signal of 106 electrons per pixel for 1 second of exposure time, from eq.(3.26) (see Chapter 3) the shot noise would be 103 electrons. This is the maximum shot noise provided by the CCD.
In the instrument described in this chapter the total noise of the CCD was mainly dominated by the shot noise contribution when using a laser and a halogen light source. In contrast, for Raman signals shot noise may not be the dominant source of noise. From eq.(3.25) for 1 s exposure time the ther-mal noise is 4.10−2 electrons per pixel. For much longer exposure times and operation at higher temperatures the dark noise might increase considerably.
Finally, the maximum readout noise for the DV420-OE CCD is 10 electrons.
In conclusion, in the DV420-OE CCD the dominant noise contribution is the shot noise when the current of photons detected at the CCD is high enough.
Otherwise, it will depend on factors such as temperature or exposure time.
The optimal CCD camera for this prototype needs to have a fast frame rate.
In the design the camera was triggered by the laser reference signal. For a fine control of the instrument the camera should be able to produce a frame each time it is triggered by a maximum of the TTL signal. Therefore, it needs to respond to frequencies in the range of 18.7 Hz-3.8 kHz. This means
that with the current step-motor-driven mirror a camera that records at least 18.7 fps is needed for a mirror with a constant velocity of 5000 nm(s−1). The available CCD was not fast enough as it reached a maximum of 3 fps at full read-out.
Two approaches can be taken to overcome the mentioned problem. In first place, a slower step-motor can be used coupled to the current CCD camera.
This will increase acquisition time and consequently contradict one of the main objectives of the prototype design, which was to speed-up the pro-cess. A second option is to use a faster detector. Fourier transform Raman systems with unidimensional high-sensitivity detectors have been widely de-veloped [130]. However, these detectors do not allow 2D imaging.
The main trade-off of common CCDs such as the one reported in this study when compared with fast point-like detectors such as photomultiplier tubes (PMT) is the higher time-response of 2D systems. In general, a faster res-ponse would have to be a compromise over sensitivity or vice versa. Ultra-fast CCDs (around thousands of fps) coupled to intensifiers that allow single photon detection are commercially available.
For the future system in the NIR the problem of etaloning is likely to ap-pear. Etaloning is an intrinsic problem of thin back illuminated CCDs which becomes of great importance in the NIR. Etaloning could represent up to the 20% of the detected radiation, masking the Raman signal. Reflections between the parallel front and back surfaces of these CCDs cause them to act as “etalons”. This behavior leads to unwanted interference fringes.