1. Introduction

Near-infrared spectroscopy is a powerful technology that enables the identification and classification of physical materials through changes in a sample's absorption or emission of light of different wavelengths. Handheld spectrometers operating at wavelengths between 700 and 2500 nm can be used in industries such as food, pharmaceutical, oil and gas, medical, security and other emerging industries. DLP-based spectrometers use a DMD and a single point detector instead of a traditional linear array detector for wavelength detection. As shown below.

By scanning the columns in sequence (turning on a specific column of pixels), light of a specific wavelength is directed to the detector and captured. DLP technology in near-infrared (NIR) spectroscopy offers the following advantages:

1. Higher performance is obtained by using a larger single-point 1mm detector compared to linear detectors, which only have small pixel acquisition capabilities.

2. By using single-point InGaAs detectors and low-cost optics, system costs can be reduced. And high-resolution DMD digital micromirrors allow custom patterns to compensate for the optical distortion of each individual system.

3. The captured signal is stronger, not least because DMD can use fast, flexible and programmable pattern and spectral filters compared to traditional DMD.

4. Through programmable mode, the DLP spectrometer can:

a) Change the light intensity of the detector by controlling the number of pixels in a column.

b) Change the resolution of the system by controlling the width of the columns.

c) Use a set of Hadamard patterns, each pattern can capture multiple wavelengths of light, and then retrieve the wavelengths through a decoding process. Each pattern turns on 50% of the DMD pixels at a time, directing a much larger signal into the detector than a column scan.

d) Use custom spectral filters to select specific wavelengths of interest.

The DLP near-infrared spectrometer is a rear dispersion architecture with a removable reflection sample. The unit reflection module includes two lens-end broadband tungsten lamps. Place a sample on the reflective head. During scanning, the sample absorbs a specific amount of near-infrared light and diffusely reflects the unabsorbed light into the system. The amount of light absorbed at each wavelength depends on the material's molecular makeup and is the material's unique chemical fingerprint. The diffusely reflected light sample is collected by the collection lens and focused into the optical engine through the input slit. The slit size is chosen to balance wavelength resolution with the SNR of the spectrometer.

For example, when designing a 900-1700nm spectrometer, the ZMAX optical path simulation diagram is as follows. A slit of 25µm wide by 1.8 mm high is used. The slit is collimated by the first set of lenses, passes through an 885nm long wavepass filter, and then is incident. Reflection grating. This grating is combined with a focusing lens to disperse the light to the corresponding wavelength position. The focusing lens forms an image of the slit at the DMD digital micromirror. Different wavelengths of the slit image spread horizontally across the DMD. The optical system images the 900nm wavelength at one end of the DMD, the 1700nm wavelength at the other end, and all other wavelengths dispersed sequentially in between. When a specific DMD column is selected to be open or tilted to a certain positive angle, the energy reflected by the selected column is directed through the collection optics to the single-pixel InGaAs detector. All other DMD columns are selected to be closed or tilted to some negative angular position, shifting unselected wavelengths to the bottom of the light engine and away from the detector optics so as not to interfere with selected wavelength measurements. Taking into account mechanical tolerances in slit position, grating angle and DMD position, the slit image on the DMD is 10% unfilled at each end along the dispersion axis. Calibration is required during the manufacturing process between wavelengths and their column positions on the DMD. The DMD column is usually not divisible by the number of required wavelength groups, and the column width needs to be kept constant during the scan. The amount of steps depends on the desired width and number of patterns (wavelength points), and the step size can vary from the column width.

The wavelength in DLP corresponds to the quantized pixel at the center of the column width. Since different DLP spectrometers have different calibration parameters, the wavelengths vary from spectrometer to spectrometer. Wavelength interpolation can use the spline-interpolation interpolation algorithm to oversample the input wavelength vector to a common wavelength vector, interpolating different data sets to a common wavelength position vector before comparison, and then use Svitsky Golay's smoothing algorithm.

The DLP reflection module works by illuminating the sample under test at an angle, not collecting specular reflections, while collecting and focusing diffuse reflections into the slit. This illuminator is designated a lens-end lamp because the front end of the glass bulb directs more light from the filament to the lens in the sample test area. The figure below shows the top view of the lighting module. The top of the illustration depicts the cavity of the slit. The bottom inset depicts the blue sample window. The green rectangle represents the lens end light. This dark yellow cone is the light output by the lamp. Each lamp produces a 40-degree beam with a 0.75mm high crossover across the blue window. Consideration needs to be given to mechanical tolerances of the chassis, lens tip to lamp to lamp, lamp shape variations, and lamp placement. The lens-end lamp focuses the beam approximately 3 mm from the lamp and produces a spot size that covers the blue sample window.

2. Hardware part

The hardware architecture diagram is as follows. Among them, the amplifier noise is usually modeled as an output noise spectral density plot, from which the total integral can be obtained to calculate the RMS noise value. Amplification is usually provided by a transimpedance amplifier (TIA) in photoconductive or photovoltaic mode. In most applications, a photovoltaic mode TIA will provide the lowest noise analog front end. If there is a need to sample very quickly (faster than 100KHz), photoconductive methods may be required. The designed bandwidth should be high enough to allow the signal to fully stabilize during the pattern exposure period. If the bandwidth is lower than this, inaccurate spectra will be produced. If the bandwidth is much higher, the ADC may have to be faster and more expensive in order to filter out the noise before input to the ADC.

The noise at the detector's current output is a function of the optical power at the detector, with respect to the detection rate and the area of ​​the detector. This includes Johnson noise, shot noise, and dark noise. Since noise increases with detector area, it is beneficial to focus the light from the DMD so that the convergence point can be as small as possible. As the effective area of ​​the DMD increases, larger detectors may be required. Detector noise can be reduced by cooling the detector with TEC. Generally, less power is required to cool a single point detector to a given temperature than to cool a ratio array detector because less mass needs to be cooled.

Light source stability is also very important for units designed with integrated light sources. Any noise source will appear as noise at the detector, potentially exhibiting errors in the output spectrum. Possible drive methods include:

1. Constant voltage

This method is usually the cheapest, but has a major drawback: any change in contact resistance, a change in the lamp wire resistance, will cause a change in the drive current. Therefore under these conditions the brightness of the lamp is no longer constant.

2. Constant current

This is usually the preferred method due to its simplicity. Current monitoring can also be performed via the sensing resistor. Another reason why a current source is preferred when driving an incandescent lamp is that the resistance of the filament changes dramatically as the lamp heats up.

3. Wavelength**** calibration

Accurate and stable calibration, including wavelength correction and radiometric correction, is key to spectrometer design. We assume that the main sources of optically or mechanically induced distortion can be restricted to second-order 2D polynomials.

Before calibration, there may be unknown wavelength positions, distortions and rotations on the DMD, as shown in the figure below (distortion amplification).

For wavelength calibration, we illuminate the slit peak with a spectrum of known emission or absorption. We can then scan with a small rectangular area of ​​the DMD mirror turned on and then scan the mirror block in small increments on the DMD as shown in the image below.

These status pixel rectangles should have the following properties:

1. Height

This should be short enough that the desired distortion is relatively minimal in this region. Typically, 1/5 to 1/9 of the non-dispersive size of the DMD is used.

2. Width

This should be narrow enough to get accurate peak positions from each calibration peak, but wide enough to get enough signal to reduce noise. In practice, the rectangle of this state pixel should be the same width or narrower than the image of the slit on the DMD. Scanning this pattern across the DMD centered on the top, middle, and bottom rows of the DMD produces three spectra. A peak finding algorithm should then be used to locate the DMD column corresponding to the known wavelength peak of the calibration source.

Regarding wavelength correction, this article describes how to implement wavelength correction based on second-order polynomial coefficients and using the wavelength corresponding to the known DMD position. First, construct a coefficient vector , which determines that we can obtain the wavelength through the DMD position , which satisfies formula (1). Obtained through the least squares method . Then, based on the actual DMD position we can calculate it.