Method and device for realizing accurate measurement of molecular concentration by applying singular coefficient analysis technology

Abstract

The invention provides a method and a device for realizing accurate measurement of molecular concentration by applying a singular coefficient analysis technology. The method comprises the following steps: generating a time domain absorbance signal by combining a high-performance high-intensity monochromatic pulse emitting laser with a tip ultra-high-speed microelectronic technology; further, an advanced artificial intelligence algorithm is utilized to directly implement data mining on the time domain absorbance signal, and the molecular concentration is accurately measured. The method subverts the field of traditional spectroscopy, abandons a method of converting an absorbance signal into a molecular absorption spectrum and a spectral peak, and directly converts the absorbance signal into a singular spectrum and a key singular coefficient which have a new look for a spectroscope, so that information loss is greatly reduced, and the speed and the accuracy of molecular analysis are remarkably improved. Compared with a traditional Fourier transform infrared spectrometer, the device is better in overall performance, the size and the weight of the device are greatly reduced, and a new path is opened up for innovative research and development of portable and even wearable molecular analyzers.

Description

Technical Field

[0001] The present invention relates to the field of quantitative measurement of general molecular concentrations, and more specifically, to the use of an artificial intelligence optimized quantum cascade laser device system to analyze key singular coefficients through innovative methods to achieve high-speed and accurate measurement of molecular concentrations. Background Art

[0002] Analytical chemistry occupies an important academic position, focusing on the confirmation of the identity of chemical substances and the quantitative determination of their concentration. Every atom and molecule in the chemical world has a unique electronic structure, which is constructed by a series of discrete energy states. When light is irradiated onto a chemical atom or molecule, it will induce its transition from one discrete energy state to another discrete energy state, thereby absorbing light of a specific wavelength. The graph of light intensity changing with wavelength is the atomic spectrum or molecular spectrum we are familiar with. They are like the unique "fingerprints" of chemical substances and are widely recognized and applied in chemical analysis. Furthermore, through the precise quantitative analysis of the absorbance, the accurate concentration value of the chemical substance can be obtained. Spectroscopy, a scientific and technological technology that deeply explores the absorption spectra of atoms and molecules, has undoubtedly laid a solid foundation for the development of analytical chemistry.

[0003] In many fields with far-reaching impact on the social economy, such as petrochemicals, polymer manufacturing, medical diagnosis, drug research and development and production, food production and processing, and environmental protection, rapid and accurate measurement of molecular concentrations is crucial, and it is best to be done with the help of small and portable instruments. Because the molecules in these industries are usually polyatomic, there are many types of chemical bonds arranged at different bond angles. Distinguishing them in chemical analysis can no longer simply rely on the differences in their absorption characteristics in terms of electronic excitation in the x-ray or UV-visible range, because the spectral peaks associated with these excitations are not sharp enough to resolve the composition and conformation of polyatomic molecules. In fact, the best way to analyze these molecules is to measure the infrared absorbance of each molecular fingerprint as a function of wavenumber. Each absorbance signal is related to infrared-allowed vibrational excitations, and is related to a specific mode of bond stretching, bond bending, or a combination of them. In fact, infrared spectroscopy has been developed for more than a century, and Fourier transform infrared (FTIR) spectrometers have become the most popular instruments for molecular analysis. Generally speaking, the most advanced FTIR spectrometers, such as the Nicolet iS50 FTIR spectrometer, include a thermal infrared lamp for luminescence and a Michelson interferometer for infrared monochromatization, which displays the absorbance signal in the time domain through the Fourier transform of the interference pattern. However, the size of a Fourier transform infrared spectrometer is like a suitcase, weighing about 70 kg, which is not convenient to use. Although many attempts have been made to reform and miniaturize infrared molecular analyzers, including FTIR spectrometers, it was not until the invention of quantum cascade lasers (QCLs) that great improvements were made. QCLs are semiconductor infrared lasers that are brighter and smaller than the infrared lamps of traditional infrared spectrometers. Currently, the single-mode QCLs on the market have highly monochromatic laser beams with a broadband of more than 200cm -1 Multimode QCLs, as well as broadband QCLs coupled with tunable spectral filters, have been used to develop novel molecular analyzers. Clearly, QCL-based molecular analyzers have many advantages over state-of-the-art FTIR spectrometers, such as being more sensitive, faster, more accurate, smaller, and lighter. In addition to improving the performance of FTIR spectrometers by replacing infrared lamps with QCLs, successful cases of redesigning and redefining FTIR spectrometers have also emerged.

[0004] However, all the innovations of new molecular analyzers, including those with QCL, still follow the basic principle of spectroscopy, that is, for molecular analysis, first plot the infrared molecular spectrum and then calculate the molecular concentration. This spectral analysis feature is faithfully enhanced in the operation of FTIR spectrometers, because Fourier transform procedures are used to convert each interference pattern recorded by the spectrometer's Michelson interferometer into an infrared molecular spectrum. To date, no one has carefully examined whether molecular concentration data can be directly data-mined from the interference pattern containing molecular absorbance information to suppress any possible information loss and distortion. Just as the FTIR spectrometer operates to convert the interference pattern into a molecular spectrum, QCL-based molecular analyzers usually also include the absorbance signal of the monochromatic QCL-pulses arranged in the time domain, and all these QCL-based innovations blindly follow the FTIR spectroscopy method to convert this time-series absorbance signal into an infrared molecular spectrum in the time domain. The present invention proves with conclusive evidence of experimental results that indeed this drifting situation has largely hindered the development of molecular analysis including QCL-based. There are other disadvantages of molecular spectroscopy methods. Typically, a molecular absorption spectrum consists of multiple molecular absorption spectrum peaks. This diversity brings trouble to molecular spectroscopy analysis because its concentration results depend on the precise algorithm of peak selection and on the molecular absorption spectrum interference from the components present in the test sample. The loss of accuracy due to this shortcoming of molecular spectroscopy analysis methods has recently been studied and confirmed.

[0005] Considering the various limitations of traditional FTIR spectroscopy, it is clear that the existing spectral molecular analysis methods are in urgent need of innovation. This requires not only the introduction of advanced hardware equipment to overcome the size and weight deficiencies of FTIR spectrometers, but also the use of cutting-edge software technologies including artificial intelligence to comprehensively improve the speed and accuracy of quantitative molecular analysis.

[0006] In short, the once all-powerful FTIR spectroscopy is facing a strong impact from emerging technological advances, such as QCL in the hardware field and artificial intelligence data management in the software level, which are challenging the advantages of FTIR in the field of quantitative molecular analysis. The market demand for compact, fast and accurate molecular analyzers to achieve quantitative molecular concentration measurement is growing, and this demand has not been fully met in the current technology system. In response to these needs, innovation should abandon the constraints of traditional spectroscopy categories and established principles, and be brave to explore and develop new solutions. Summary of the invention

[0007] In view of the problems existing in the prior art, a method and device for realizing accurate measurement of molecular concentration using singular coefficient analysis technology are provided. By analyzing key singular coefficients through innovative methods, high-speed and ultra-precise measurement of molecular concentration can be realized.

[0008] A method for accurately measuring molecular concentration using singular coefficient analysis technology, the method comprising: directly mining the time-domain absorbance signal of a test sample to obtain a singular base and a singular spectrum, then constructing a set of key singular coefficients, and then performing data mining on the molecular concentration of the test sample from multiple key singular coefficients.

[0009] In one embodiment of the present application, the test sample is gas, liquid or solid.

[0010] In one embodiment of the present application, the time-domain absorbance signal comprises the fingerprint absorbance characteristics of the molecular components in the test sample.

[0011] In one embodiment of the present application, the time-domain absorbance signal includes an absorbance signal measured as a function of time, and the absorbance signal changes continuously in a time series according to a change in the wave number of the monochromatic light over time.

[0012] In one embodiment of the present application, the process of directly data mining the time-domain absorbance signal of the test sample to obtain singular cardinalities and singular spectra includes four consecutive steps of multifractal detrended fluctuation analysis and singular coefficient analysis similar to multifractal detrended fluctuation analysis, and the four consecutive steps include the step of calculating a filtered absorbance signal profile from the time-domain absorbance signal, the step of calculating a variance function from the filtered absorbance signal profile, the step of calculating a series of q-order fluctuation functions from the variance function, and the step of calculating singular cardinalities and singular spectra from a series of q-order fluctuation functions.

[0013] In one embodiment of the present application, the step of calculating the filtered absorbance signal profile from the time-domain absorbance signal comprises taking the absorbance x of the i-th time unit in the time-domain absorbance signal i , calculate its average value when the total duration is N time units <x>, and construct the filtered absorbance signal profile Y(i):

[0014]

[0015] In one embodiment of the present application, the step of calculating the variance function from the filtered absorbance signal profile Y(i) includes dividing the signal profile Y(i) from i=1 to i=N into M non-overlapping segments of equal time length s, and then dividing it from i=N to i=1 into another group of M non-overlapping segments of time length s, and then calculating the local trend y of the vth segment using the least squares fitting of the polynomial function. v (i) and determine the variance function Y of the vth segment var (v,s):

[0016]

[0017] In one embodiment of the present application, the step of calculating a series of q-order fluctuation functions from the variance function includes calculating a series of q-order fluctuation functions F(q,s):

[0018]

[0019] In one embodiment of the present application, the step of calculating key singular coefficients from a series of q-order wave functions includes calculating the singular cardinality β, the singular intensity α, and the multifractal dimension f(α) to construct a singular spectrum f(α), wherein h(q) is calculated by d(log(F)) / d(log(s), h′(q) is calculated by d(h) / d(q),

[0020]

[0021] α=h(q)+qh′(q)

[0022] f(α)=q 2 h′(q)+1

[0023] In one embodiment of the present application, constructing a set of key singular coefficients includes: extracting three singular intensities consisting of the leftmost point, the rightmost point and the maximum point from the obtained singular spectrum to form a first-level key singular intensity, and deriving multiple second-level key singular intensity from the first-level key singular intensity; the singular cardinality, the first-level key singular intensity and the second-level key singular intensity constitute a set of key singular coefficients.

[0024] In one embodiment of the present application, the plurality of secondary key singular intensities include a secondary key singular intensity defined as the width of the leftmost point and the rightmost point, and, if necessary, other derivative values ​​of the primary key singular intensity.

[0025] In one embodiment of the present application, the data mining of the molecular concentration of the test sample from multiple key singular coefficients includes: training a machine learning regression model using multiple key singular coefficients of several reference samples with known molecular concentrations, and determining the unknown molecular concentration of the test sample by projecting the key singular coefficients of the test sample onto the trained machine learning regression model.

[0026] In one embodiment of the present application, the reference sample comprises a sample having the same properties as the test sample, except that the concentration of the molecule to be detected is known.

[0027] In one embodiment of the present application, the machine learning regression model includes a standard support vector machine regression model, or any other machine learning regression model.

[0028] In a second aspect, the present application proposes a device for realizing precise measurement of molecular concentration by using singular coefficient analysis technology, the device comprising a laser system, a light-sample interaction unit, a detector, an analog-to-digital converter and a main control unit, wherein the laser system is used to generate the required laser pulses, the light-sample interaction unit is used to complete the reaction between the laser pulses and the test samples, the detector is used to detect the reaction results in the light-sample interaction unit, the analog-to-digital converter is used to convert the detection results of the detector into analog-to-digital and output them to the main control unit, the main control unit comprises hardware programmed with software to act as a singular coefficient calculator and a machine learning regression model processor, and is used to realize precise measurement of molecular concentration according to the output results of the analog-to-digital converter.

[0029] In one embodiment of the present application, the main control unit includes an integrated circuit chip, a plurality of integrated circuit chips packaged using surface mount technology on a printed circuit board, a programmable logic controller, a microcomputer or a computer.

[0030] In one embodiment of the present application, the main control unit includes multiple connection circuits connected to the laser system for controlling the laser system; multiple connection circuits connected to the detector for controlling the detector; and multiple connection circuits connected to the analog-to-digital converter for controlling the analog-to-digital converter.

[0031] In one embodiment of the present application, the main control unit is connected to a display for displaying the molecular concentration measured by the device.

[0032] In one embodiment of the present application, the main control unit includes a telecommunication device for transmitting the molecular concentration measured by the device to a plurality of receivers.

[0033] In one embodiment of the present application, the main control unit comprises a telecommunication device for transmitting the molecular concentration measured by the device to a mobile phone.

[0034] In one embodiment of the present application, the singular coefficient calculator includes hardware programmed by software for data mining singular bases and singular spectra from the absorbance signal output by the analog-to-digital converter in the time domain, and constructing a set of key singular coefficients.

[0035] In one embodiment of the present application, the machine learning regression model processor includes hardware programmed with software for data mining molecular concentrations from a set of key singular coefficient outputs output by the singular coefficient calculator.

[0036] In one embodiment of the present application, the machine learning regression model processor includes hardware with software for setting up and applying a support vector machine regression model to data mine molecular concentrations from a set of key singular coefficients output by the singular coefficient calculator.

[0037] In one embodiment of the present application, the laser system includes a broadband laser and a spectral filter for emitting time-domain monochromatic pulses, and the wave number of the pulses arranged in the time domain also changes sequentially with time.

[0038] In one embodiment of the present application, the broadband laser is a QCL (quantum cascade laser).

[0039] In one embodiment of the present application, the QCL includes a direct or indirect connection to the main control unit to receive and track control signals from the main control unit, and then emit laser pulses with a frequency of 10 kHz to 1 MHz in the time domain.

[0040] In one embodiment of the present application, the QCL includes a signal directly or indirectly connected to the main control unit to receive and track control signals from the main control unit, thereby emitting laser pulses with a frequency of 100 kHz to 500 kHz in the time domain.

[0041] In one embodiment of the present application, the QCL includes a signal directly or indirectly connected to the main control unit to receive and track control signals from the main control unit, thereby emitting laser pulses with a frequency of 150 kHz to 200 kHz in the time domain.

[0042] In one embodiment of the present application, the QCL comprises from 20 to 300 cm -1 bandwidth.

[0043] In one embodiment of the present application, the QCL comprises a wavelength from 40 to 200 cm -1 bandwidth.

[0044] In one embodiment of the present application, the QCL comprises about 200 cm -1 bandwidth.

[0045] In one embodiment of the present application, the spectral filter includes a plurality of connection circuits connected to the main control unit, and is used to receive control instructions from the main control unit to correctly operate as a spectral filter.

[0046] In one embodiment of the present application, the spectral filter comprises a grating, the rotation of which is controlled by a micro-electromechanical actuator, and the spectral resolution of the spectral filter is controlled by a plurality of apertures along the optical path of the laser.

[0047] In one embodiment of the present application, the spectral filter comprises a grating whose length and filtering wavenumber are compactly controlled by voltages applied to two end electrodes of the grating plate, and the spectral resolution of the spectral filter is controlled by multiple apertures along the laser optical path.

[0048] In one embodiment of the present application, the spectral resolution is set to 0.05-4 cm -1 .

[0049] In one embodiment of the present application, the spectral resolution is set to 1-4 cm -1 .

[0050] In one embodiment of the present application, the spectral resolution is set to 1.6 cm -1 .

[0051] In one embodiment of the present application, the laser pulse width filtered by the spectral filter is set to 50-500ns.

[0052] In one embodiment of the present application, the laser pulse width filtered by the spectral filter is set to 50-200ns.

[0053] In one embodiment of the present application, the laser pulse width filtered by the spectral filter is set to 50-100ns.

[0054] In one embodiment of the present application, the laser pulse width filtered by the spectral filter is set to 100ns.

[0055] In one embodiment of the present application, the light-sample interaction unit comprises an infrared transparent unit for accommodating the test sample in a gas, liquid or solid state.

[0056] In one embodiment of the present application, the light-sample interaction unit comprises an infrared transparent single attenuated total reflection prism.

[0057] In one embodiment of the present application, the light-sample interaction unit comprises an infrared transparent multiple attenuated total reflection prism.

[0058] In one embodiment of the present application, the detector comprises an InAsSb infrared detector.

[0059] In one embodiment of the present application, the detector comprises an MCT infrared detector.

[0060] In one embodiment of the present application, the analog-to-digital converter includes an analog-to-digital converter operating at a sampling frequency of 10-100 MHz.

[0061] In one embodiment of the present application, the analog-to-digital converter includes an analog-to-digital converter operating at a sampling frequency of 20-60 MHz.

[0062] In one embodiment of the present application, the analog-to-digital converter includes an n-bit analog-to-digital converter with n ranging from 10 to 24.

[0063] In one embodiment of the present application, the laser system, the light-sample interaction unit, the detector, the analog-to-digital converter and the main control unit are packaged on a plastic printed circuit board or a ceramic printed circuit board using surface mounting technology.

[0064] In one embodiment of the present application, the laser system, the light-sample interaction unit, the detector, the analog-to-digital converter and the main control unit are all integrated into a single chip.

[0065] In one embodiment of the present application, the device is designed as a portable molecular analyzer, which is more than 10 times smaller and more than 10 times lighter than a typical FTIR spectrometer.

[0066] In one embodiment of the present application, the device is designed as a wearable molecular analyzer, which is more than 100 times smaller and more than 100 times lighter than a typical FTIR spectrometer.

[0067] Some aspects of the present invention currently disclosed represent a revolutionary breakthrough in the technology of molecular concentration measurement. Traditionally, the field has adhered to a dogmatic approach, which is to directly convert the raw data of molecular light absorption, whether in the time domain or the frequency domain, into a molecular absorption spectrum, and then deduce the molecular concentration by analyzing the intensity regression of the peak position in the molecular absorption spectrum. However, this approach has now been questioned. More specifically, the embodiments of the present invention have a deep insight into the fact that QCL-based molecular analyzers are increasingly occupying a core position by virtue of the use of raw molecular absorption data in the time domain. To this end, cutting-edge artificial intelligence technology is introduced to efficiently extract molecular concentration information directly from the raw molecular absorption data in the time domain. This innovation gives the embodiments of the present invention a significant advantage in the field of ultrafast and ultra-precise molecular concentration measurement, effectively reducing the time loss and information loss in the data processing and conversion process.

[0068] The molecular concentration analyzer of the present invention draws on the instrument design of a miniaturized infrared molecular analyzer and adopts a compact and lightweight semiconductor laser system. The system can output high-intensity narrow-bandwidth pulses in the time domain to generate raw data of molecular absorption. These strong narrow-bandwidth pulses in the time domain are coupled with the light-sample interaction unit to ensure that the target molecules in the test sample can effectively absorb light and generate corresponding absorbance signals. In addition, the analyzer is also equipped with an ultrafast detector and an ultrafast analog-to-digital converter (ADC), which work together to accurately measure the molecular absorption coefficient and provide a digital molecular absorbance signal in the time domain. The instrument in the current technology follows the spectral data management protocol, which covers the process of converting the time-domain comb absorption pulse into a molecular absorption spectrum, and performing regression analysis on the molecular absorption spectrum peak to estimate the molecular concentration. In contrast, the embodiment disclosed in the present invention uses a main control unit (MCU) to control the laser system, detector and ADC, and directly performs data mining on the time-domain digitized absorbance signal obtained by the ADC to generate an estimation result of the molecular concentration. In addition, existing technical instruments that follow the spectral data management protocol often rely on computer assistance to achieve precise control of laser systems, detectors, ADCs and their ancillary equipment (such as signal generators and other control units). At the same time, the computer also assists spectrometer users in selecting the molecular absorption spectrum peaks that are most suitable for molecular absorption spectrum analysis and concentration estimation. Compared with traditional methods, the embodiments disclosed in the present invention cleverly use MCU to coordinate the control and data management of all instruments, aiming to maximize the use of existing artificial intelligence tools while minimizing human intervention and accurately generating molecular concentration measurement results. Therefore, with the help of the implementation scheme of the present invention, concentration measurements can be completed quickly in 1 second or less, and can significantly reduce the risk of human misjudgment and errors.

[0069] The novelty of the methods and apparatus of the present invention lies in that they reject traditional spectroscopic methods and instruments, because the information and time loss caused by the conversion of raw absorbance signals into molecular spectra can be avoided by directly data mining the raw absorbance signals to generate singular spectra and a set of key singular coefficients. In addition, the methods and apparatus of the present invention generate a set of key singular coefficients from each set of raw absorbance signals, and the trained machine learning regression model contained in the methods and apparatus of the present invention gives exclusive and most accurate molecular concentrations based on the set of key singular coefficients. In contrast, traditional spectroscopic methods and instruments usually generate molecular spectra with multiple absorbance peaks and absorbance valleys from a set of raw absorbance signals of molecules, and give multiple molecular concentrations, which will be different due to differences in algorithms and human judgment for selecting and converting these absorbance peaks and absorbance valleys. Although the concepts of singular coefficient analysis, multifractal detrended fluctuation analysis, singular spectra and singular intensities have been successfully applied to predict useful information from time domain data in the fields of traffic control, weather forecasting, financial assets, health management and mechanical failure prevention, they have never been used in the field of molecular analysis. Like other multifractal analysis methods, it is also used for pattern recognition and image analysis, including analysis of reflection images and optical imaging. However, these applications are not related to the present invention. In addition, the definition, algorithm and application of the "singular cardinality" created by the present invention have not been disclosed in any existing documents including patent applications, authorized patents and published papers. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 Flow chart of a new method for directly mining the time-domain absorbance signal of a test sample to obtain the molecular concentration according to an embodiment of the present invention.

[0071] Figure 2 Schematic diagram of a novel apparatus for directly mining molecular concentrations from time-domain absorbance signals according to an embodiment of the present invention.

[0072] Figure 3 FIG. 1 is a simplified schematic diagram of a laser system according to an embodiment of the present invention.

[0073] Figure 4 It is a schematic diagram of a device in one embodiment of the present invention.

[0074] Figure 5 is a schematic diagram of an absorbance signal in one embodiment of the present invention, wherein (a) is Figure 4 Schematic diagram of the time-domain absorbance signal generated by the device, and (b) the interference pattern generated by a typical FTIR spectrometer using an interferometer.

[0075] Figure 6 It is a schematic diagram of the method disclosed in one embodiment of the present invention, including a process from generating an original time-domain absorbance signal, to direct singular coefficient analysis data mining of key singular coefficients, and finally to statistical analysis of glucose concentration.

[0076] Figure 7 Schematic diagram of 41 singular spectra in one embodiment of the present invention.

[0077] Figure 8 41 infrared molecular absorption spectra converted from time-domain absorbance signals to molecular spectra in one embodiment of the present invention.

[0078] Fig. 9 1 shows the dependence of five key singular coefficients as concentration functions in one embodiment of the present invention, wherein (a) shows the dependence of the singular base on the molecular concentration, (b) shows the dependence of the leftmost point on the molecular concentration, (c) shows the dependence of the rightmost point on the molecular concentration, (d) shows the dependence of the peak value on the molecular concentration, and (e) shows the dependence of the width of the leftmost point and the rightmost point on the molecular concentration.

[0079] Fig.10 Schematic diagrams comparing different measurement methods in one embodiment of the present invention, wherein (a) is a schematic diagram comparing the concentration measured by direct data mining and singular coefficient analysis of the original time-domain absorbance signal with the concentration of a reference blood glucose meter, (b) is a schematic diagram comparing the concentration measured by converting the time-domain absorbance signal into molecular spectroscopy with the concentration of a reference blood glucose meter, and (c) is a schematic diagram comparing the concentration measured by the new device and new method after replacing the ADC with the concentration of the reference blood glucose meter.

[0080] Fig.11 1 and 1 are singular spectra of three different albumin concentrations in one embodiment of the present invention.

[0081] Fig.12 FIG. 1 is a schematic diagram of a QCL system in an embodiment of the present invention.

[0082] Fig.13 Schematic diagram of packaging a device for accurately measuring molecular concentration using singular coefficient analysis technology on a printed circuit board in one embodiment of the present invention.

[0083] Fig.14 This is a schematic diagram of an implementation of an apparatus for accurately measuring molecular concentration using singular coefficient analysis technology integrated on a single chip in another embodiment of the present invention.

[0084] Figure numerals: 100 - laser system, 200 - light-sample interaction unit, 300 - detector and ADC assembly, 400 - main control unit, 101 - quantum cascade laser, 102 - grating plate, 103 - micro-electromechanical device, 401 - singular coefficient calculator, 402 - machine learning regression model processor. DETAILED DESCRIPTION

[0085] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar modules or modules with the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be construed as limitations on the present application. On the contrary, the embodiments of the present application include all changes, modifications and equivalents that fall within the spirit and connotation of the appended claims.

[0086] To improve the speed and accuracy of quantitative molecular analysis, please refer to Figure 1 , an embodiment of the present invention proposes a method for realizing accurate measurement of molecular concentration by using singular coefficient analysis technology, the method comprising: directly mining the time-domain absorbance signal of the test sample to obtain singular cardinality and singular spectrum, then constructing a set of key singular coefficients, and then performing data mining on the molecular concentration of the test sample from multiple key singular coefficients. Among them, the process of directly mining the time-domain absorbance signal of the test sample to obtain the singular spectrum comprises four consecutive steps of multifractal detrended fluctuation analysis and singular coefficient analysis similar to multifractal detrended fluctuation analysis, the four consecutive steps comprising the steps of calculating the filtered absorbance signal profile from the time-domain absorbance signal, the step of calculating the variance function from the filtered absorbance signal profile, the step of calculating a series of q-order fluctuation functions from the variance function, and the step of calculating the singular cardinality and singular spectrum from a series of q-order fluctuation functions.

[0087] For further information, please refer to Figure 2 An embodiment of the present invention further proposes a device for realizing precise measurement of molecular concentration by using singular coefficient analysis technology. The device includes a laser system, a light-sample interaction unit, a detector, an analog-to-digital converter, and a main control unit. The main control unit includes hardware programmed with software to act as a singular coefficient calculator and a machine learning regression model processor. Each unit and system cooperates to complete the precise measurement of molecular concentration.

[0088] In some embodiments, the laser system includes a QCL system that emits a time sequence of intense narrow bandwidth pulses in the mid-infrared wavelength range. In some embodiments, the QCL system includes a broadband QCL whose high intensity broadband laser emission is achieved by a repetitive time sequence of square wave pulses input to the QCL system and the laser emission pulses entering a spectral filter, so that the QCL system outputs an intense narrow bandwidth time sequence of pulses in the mid-infrared wavelength range.

[0089] In some embodiments, the spectral filter includes a micro-electromechanical actuator for driving a grating plate to rotate to filter broadband laser pulses into narrow-band laser pulses, and the temporal wavelength variation of the narrow-band laser pulses is controlled by the rotation angle of the grating plate, which is driven by the micro-electromechanical actuator in response to a repetitive sinusoidal wave signal input to the QCL system.

[0090] In some embodiments, the spectral filter includes a grating plate made of piezoelectric material so that the grating can be stretched with an applied voltage to filter broadband laser pulses into narrowband laser pulses, and the wavelength variation of the narrowband laser pulses in the time domain is controlled by the applied voltage.

[0091] In some embodiments, the light-sample interaction unit is an infrared transparent unit that contains the test sample in a gaseous, liquid or solid state.

[0092] In some embodiments, the light-sample interaction unit is an infrared-transparent single attenuated total reflection prism (ATR) or an infrared-transparent multiple attenuated total reflection prism (MATR), the top surface of which contains a liquid or solid test sample so as to guide the laser pulse entering the ATR or MATR into the test sample through one or more reflections; the laser pulse carrying the absorption properties of the test sample is guided to the infrared detector.

[0093] In some embodiments, ultrafast detectors paired with ultrafast analog-to-digital converters (ADCs) are fast electronic devices with response times of no more than a few nanoseconds.

[0094] In some embodiments, the MCU includes a monolithic integrated circuit chip powered by a common external power supply. The integrated circuit chip has an intelligent software program that receives digitized narrow-band laser pulses with molecular absorption characteristics, performs multi-fractal detrended fluctuation analysis or similar singular coefficient analysis on the time-domain absorbance signal and generates a singular spectrum, and calculates a set of key singular coefficients. A machine learning regression model is constructed using multiple sets of key singular coefficients from multiple molecular measurements, and the molecular concentration of the unknown test sample is calculated using the machine learning regression model and the key singular coefficient set generated by the molecular measurement values ​​of the unknown test sample. The integrated circuit chip also outputs a display of the molecular concentration of the sample being tested, and outputs electronic control signals to drive the laser system, detector and ADC to work normally accordingly.

[0095] In some embodiments, the MCU includes a dedicated microcomputer powered by a common external power supply. The microcomputer has an intelligent software program that receives digitized narrow-band laser pulses with molecular absorption characteristics, performs singular coefficient analysis on the time-domain absorbance signal and generates singular bases and singular spectra, and calculates a set of key singular coefficients. A machine learning regression model containing multiple sets of key singular coefficients is constructed using data from multiple molecular measurements, and the molecular concentration of the unknown test sample is calculated using the key singular coefficient set generated by the model and the molecular measurement data of the unknown test sample.

[0096] In some embodiments, the microcomputer outputs a display of the molecular concentration of the sample being measured and outputs electronic control signals to drive the correct operation of the laser system, detector and ADC.

[0097] In some embodiments, the microcomputer outputs electrical signals to an external display device to display the molecular concentration of the test sample, and outputs electrical signals to an external electrical signal generator to drive the correct operation of the laser system, the detector and the ADC.

[0098] like Fig.13 As shown, the QCL laser system 100, the light-sample interaction unit 200, the detector and ADC assembly 300 and the MCU 400 of the device are all packaged on a printed circuit board using surface mount technology to provide a compact version of the device to practice the present invention.

[0099] like Fig.14 As shown, all components of the device including the QCL laser system 100, the light-sample interaction unit 200, the detector and ADC assembly 300 and the MCU 400 can be integrated into a single chip.

[0100] In some embodiments, the singular coefficient analysis comprises the following calculation steps:

[0101] (1) Take the absorbance x of the i-th time unit in the time domain absorbance signal i , calculate its average value when the total duration is N time units <x>, and construct the filtered absorbance signal profile Y(i),

[0102]

[0103] (2) The signal Y(i) is divided into M non-overlapping segments of equal time length s from i = 1 to i = N, and then divided into another group of M non-overlapping segments of time length s from i = N to i = 1. Then, the local trend y of the vth segment is calculated using the least squares fitting of the polynomial function. v (i) and determine the variance function Y of the vth segment var (v,s),

[0104]

[0105] (3) Calculate the qth order wave function F(q,s),

[0106]

[0107] (4) Calculate the singular cardinality β, the singular intensity α, and the multifractal dimension f(α) to construct the singular spectrum f(α), where h(q) is calculated by d(log(F)) / d(log(s), and h′(q) is calculated by d(h) / d(q).

[0108]

[0109] α=h(q)+qh′(q)#(7)

[0110] f(α)=q 2 h′(q)+1#(8)

[0111] (5) Extract three first-order key singular intensities from the singular spectrum, the leftmost point α +∞ , the rightmost point α –∞ , the peak point of α, thereby constructing a set of key singular intensities. Multiple secondary key singular intensities can be derived from the primary key singular intensity, wherein the multiple secondary key singular intensities include a secondary key singular intensity defined as the width of the leftmost point and the rightmost point, and other derivative values ​​of the primary key singular intensity when necessary; the singular cardinality, the primary key singular intensity and the secondary key singular intensity constitute a set of key singular coefficients.

[0112] In some embodiments, the machine learning regression model comprises a standard support vector machine regression model.

[0113] In combination with the specific device implementation process and the operation flow of the method, the scheme for obtaining the molecular concentration by directly mining the time-domain laser absorbance signal detected from the test sample through singular coefficient analysis proposed in the present invention is further described below.

[0114] Specifically, the device ( Figure 2 ) of the laser system 100 emits a time sequence of high-intensity monochromatic laser pulses. In one complete cycle of this high-intensity monochromatic laser pulse time sequence, the laser pulses are repeatedly output in a sequence of increasing wave numbers and a sequence of decreasing wave numbers to cover a specific wave number range within which the laser placed in the device ( Figure 2 The target molecules in the test sample in the light-sample interaction unit 200 of the device will absorb some of the photons in the laser pulse. Many full cycles of this time series of pulses are emitted from the laser system 100, and the absorbance signal is generated by the device ( Figure 2 ) is detected and digitized by a detector and analog-to-digital converter (ADC) assembly 300. The laser system 100, the detector and the ADC assembly 300 are composed of the device ( Figure 2 ) is controlled by a main control unit (MCU) 400. After receiving the full-cycle time-domain digitized absorbance signals from the detector and ADC component 300, the MCU 400 directly data mines and extracts the molecular concentration of the sample under test from these time-domain signals without converting the time-domain absorbance signals into a conventional molecular absorption spectrum in a specific form with absorbance as a wavenumber function. Specifically, the MCU 400 includes a built-in circuit and control sequence for performing singular coefficient analysis on the absorbance signal in the time domain, and the hardware and software component is referred to as a singular coefficient calculator 401. The singular coefficient calculator 401 uses the time-domain absorbance signals from the detector and ADC component 300 to generate a singular spectrum and calculate a set of key singular coefficients. The singular coefficient calculator 401 then passes the set of key singular coefficients to a hardware and software component referred to as a machine learning regression model processor 402. After receiving the key singular coefficients of the reference sample with known molecular concentration, the machine learning regression model processor 402 trains a machine learning regression model for measuring the concentration of the molecule of interest, and uses the trained model to determine the molecular concentration of the test sample based on the key singular coefficients of the test sample. Finally, the MCU 400 outputs the molecular concentration of the test sample to a data display unit or remotely transmits the molecular concentration to a mobile phone.

[0115] In some embodiments, the laser system 100 includes a QCL system covering a bandwidth of several hundred wave numbers to be able to detect high-impact medical and industrial molecules whose absorption characteristics are within this bandwidth. Specifically, the QCL system emits thousands of highly monochromatic pulses in a complete cycle of tens of milliseconds, and these pulses are distributed over a bandwidth of several hundred wave numbers. In each full cycle of this emission, the detector and ADC component 300 cooperate to provide a time-domain absorbance signal at a high-speed sampling rate within a few nanoseconds, and output millions of time-domain absorbance data points to the MCU 400 in each full cycle. This sufficiently large set of time-domain data points is transmitted to the singular coefficient calculator 401 for accurately calculating the singular spectrum and a set of key singular coefficients. Then, the machine learning regression model processor 402 calculates the molecular concentration based on the set of key singular coefficients.

[0116] In some illustrative embodiments, laser system 100 includes a QCL system, such as the QCL assembly L14890-09 sold by Hamamatsu Photonics, which covers 935-1128 cm -1 Specifically, the structure of the QCL system is as follows: Figure 3 As shown, a total of 3600 pulses are emitted in a full period of 20 ms, the pulse width is 100 ns, and the wave number of each monochromatic pulse is about 0.10 cm -1 The spectral filtering in QCL-L14890-09 is as follows Figure 3 As shown, the micro-electromechanical device 103 is used to drive the rotating grating plate 102 to reduce the broadband to about 200cm -1 , broadband time series laser pulses with a pulse width of about 0.2 ms are filtered to a bandwidth of about 0.1 cm -1 The QCL-L14890-09 system is a highly monochromatic time-series laser pulse with a time width of about 100ns. The wave number of the monochromatic laser pulse increases sequentially with the change of the grating angle in the first half of the cycle and decreases sequentially in the subsequent half cycle. The laser output of the QCL-L14890-09 system is rated at 900mW, so it is much brighter than the mid-infrared intensity of any commercial FTIR spectrometer. The spectral resolution of the QCL-L14890-09 system is 1.6cm -1 , which is comparable to the spectral resolution of commercial FTIR spectrometers in normal operation. This spectral resolution can be further improved to match the best FTIR spectrometers by reducing the aperture diameter from the laser to the grating and the exit aperture of the monochromatic beam reflected by the grating.

[0117] In some illustrative embodiments, the absorbance signal is actually detected and digitally sampled by the detector and ADC assembly 300 at a sampling rate of 100 MHz. Therefore, for the entire cycle time of 20 ms per measurement, a total of 2 million data points are recorded. The singular coefficient calculator 401 and the machine learning regression model processor 402 both operate at very high speeds to actually calculate the molecular concentration of the test sample. The measurement is repeated 10 times for a test sample to further improve the measurement accuracy, and the total time is still a fraction of 1 second. As an example, the time domain absorbance pulse and singular spectrum over the entire measurement cycle are plotted at Figure 6 Obviously, both the absorption pulse and the singular spectrum in the time domain are fundamentally different from the traditional molecular absorption spectrum, which is expressed as absorbance and wave number.

[0118] In some illustrative embodiments, laser system 100 includes a QCL system, such as the QCL assembly L14890-09 sold by Hamamatsu Photonics, which covers 935-1128 cm -1 The wave number range, and Figure 4 Other components shown. Specifically, the QCL-L14890-09 system emits a total of 3600 pulses with a pulse width of 100ns in a full period of 20ms. To illustrate these pulses in the time domain, Figure 5 The first pulse with a duration of 556 μs is plotted in (a).

[0119] The detector and ADC assembly 300 with a sampling rate of 100 MHz generates a total of 2 million data points. Therefore, for each measurement of a full cycle time of 20 ms, a total of 2 million data points are recorded. The singular coefficient calculator 401 uses formulas 1-8 in the abstract section of the present invention to calculate the singular cardinality and singular spectrum of the test sample. Specifically, the number of data points N in formulas 1-2 is set to 2 million. In some illustrative embodiments, the segment length s is set to start from 160 ns and increment by 50 units of 100 ns until the maximum segment length is 5160 ns, so as to generate Y using formulas 3-5 var (v,s) and F(q,s). Using these specific singular coefficient analysis properties, the singular cardinality and singular spectra are calculated to construct a set of key singular coefficients. The machine learning regression model processor 402 uses a standard support vector machine (SVM) regression model to calculate the molecular concentration of the test sample using the key singular coefficients of the test sample and the SVM regression model trained with a reference sample of known concentration. In order to evaluate the performance of the method and device in these embodiments for actual molecular concentration measurement, patients with 41 different blood glucose concentrations were used as test samples, and three key singular coefficient measurements were performed on each test sample. Therefore, 121 sets of key singular coefficients were recorded, of which 90 were used to train the SVM regression model and 31 were used as test sets. The glucose concentration of the test sample is calculated using the trained SVM regression model and the key singular coefficients calculated from the time-domain absorbance signal of the test sample. To illustrate the measurements in these embodiments, the 41 singular spectra calculated by the singular coefficient calculator are plotted on Figure 7 The test samples are Fig.10 In order to further illustrate and clarify the difference between the singular spectrum and the conventional molecular absorption spectrum, the time-domain absorption pulses recorded by 41 test samples were converted into conventional infrared molecular absorption spectra, such as Figure 8 As shown in the figure, the blue molecular absorption spectrum at the top of the figure was recorded by FTIR spectrometer. The absorption spectrum of glucose molecules recorded by FTIR spectrometer is also included in Figure 8 middle. Figure 7 The data properties in Figure 8 The nature of the data in is quite different.

[0120] To further illustrate that the key singular coefficients are indeed well correlated with the molecular concentration, the correlation is as follows Fig. 9 As shown, Fig. 9 (a) Fig. 9 (b) Fig. 9 (c) Fig. 9 (d) and Fig. 9 (e) shows the leftmost point α of the singular cardinal numbers β and α respectively +∞ , the rightmost point α –∞ Figure 2. The peak of α and the width of the leftmost and rightmost points are correlated with the molecular concentration. While the correlations are clear, they are clearly not linear in nature. Instead, the correlations are so complex that we trained a machine learning regression model to data mine the molecular concentration from the key singular coefficients.

[0121] In order to evaluate the performance of the method and apparatus of the present invention, the blood glucose values ​​of these 41 test samples were also extracted by a standard blood glucose meter, which is considered to be the most accurate portable means of measuring blood glucose concentration. The correlation between the glucose concentration measured by this benchmark method and the glucose concentration measured using the method and apparatus of the present invention is shown on a standard Clarke error grid plot, as shown in Fig.10 (a) shows a very good correlation. In contrast, the correlation between the glucose concentration measured by the blood glucose meter and the glucose concentration measured by the apparatus of the present invention, and the correlation calculated by converting the absorption pulses in the time domain into conventional infrared molecular absorption spectra and then calculating the glucose concentration using standard spectral conventions ( Fig.10 (b)), which is obviously inferior to the correlation measured by the method and apparatus of the present invention.

[0122] In some illustrative embodiments, the methods and devices of the present invention are utilized to perform molecular analysis of albumin. Fig.11 The singular spectra of albumin at three different concentrations are given. Obviously, the three singular spectra at three different concentrations are well distributed; therefore, the method and device of the present invention are also suitable for molecular analysis of some proteins.

[0123] The conventional FTIR spectrometer and the method of the present invention are also compared in terms of their ease of operation. A typical FTIR spectrometer has a physical size more than 10 times larger than the device disclosed in the present invention, and its device structure is similar to Figure 4 The device of the illustrative embodiment shown in . Figure 4 The device described in the paper is also more than 10 times lighter than a typical FTIR spectrometer. In terms of measurement speed, repeating 10 measurements on a test sample can further improve the measurement accuracy. The total time for a typical FTIR spectroscopic operation is about a few minutes, while Figure 4 The total time of the device shown in is only a fraction of a second. Therefore, in all practical aspects, the method and apparatus disclosed in the present invention are superior to conventional FTIR spectrometers and spectrometers.

[0124] In some embodiments, the laser system 100 may refer to Figure 3 The illustrative embodiment shown in the drawings is designed and manufactured with substantial improvements in physical size and weight to the laser system 100. For example, Fig.12 The design of a quantum cascade laser 101 with a rotating grating plate 102 on a ceramic substrate with a size of 4 mm*4 mm is shown, which includes a back contact for integrating a QCL grating module for connecting with a light-sample interaction unit 200, a detector and ADC assembly 300, and an MCU 400 and is packaged on a rigid printed circuit board or a flexible printed circuit tape. In some embodiments, Fig.12 The grating plate 102 in is fabricated on a piezoelectric crystal with a large gauge factor so that an applied voltage can be used to stretch the grating and thereby tune the wave number of the emitted laser.

[0125] It is also conceivable that all components of the device of the present invention are manufactured onto a monolithic integrated module chip, which components include the laser system 100, the light-sample interaction unit 200, the detector and ADC assembly 300 and the MCU 400. Such monolithic integration will produce the most compact and accurate molecular analyzer.

[0126] The present invention will be further described below by way of examples, which are provided for illustrative purposes only and therefore should not be construed as limiting the scope of the present invention.

[0127] Example 1

[0128] like Figure 4 As shown, a commercially available QCL system, L14890-09 from Hamamatsu Photonics, is used as the -1 The exemplary device disclosed in the present invention is constructed by using a monochromatic light source with a wave number span. Figure 3 The internal structure of the exemplary light source of the exemplary device is further shown. It is obvious that the configuration of the exemplary light source is fundamentally different from the thermal infrared light source of a typical FTIR spectrometer in several aspects. First, the QCL is a semiconductor laser, which is much more energy efficient and has a photon flux that is 3-4 orders of magnitude higher than that of a thermal infrared lamp. Second, the light emitted by the QCL is coherent and highly collimated; in contrast, blackbody illumination is incoherent and poorly collimated. Therefore, the beam from the broadband QCL is more easily optically monochromated by a compact spectral filter (such as a small piece of grating tightly coupled to the QCL). Third, the laser light emitted by the QCL-L14890-09 exists in the form of a time-domain laser pulse train, and the rotation of the grating as a function of time is directly related to the change in wavenumber over time. Therefore, the monochromatic light emission of QCL is also fundamentally different from the monochromatic light emission of a typical FTIR spectrometer, because the time coordinate of the laser pulse in the time domain is linearly paired with the wavenumber coordinate of a typical molecular absorption spectrum, and the periodic laser pulse fragments arranged in the time domain can be appropriately folded together to increase the laser pulse density within the fragment until a typical infrared spectrum appears, which shows the relationship between absorbance and wavenumber. Figure 5 (a) This simple infrared spectral structure is graphically depicted, resulting from the folding of laser pulses in the time domain. In contrast, a typical FTIR spectrometer uses an interferometer to produce an interference pattern such as Figure 5 (b) shows that a Fourier transform must be performed in the wavenumber domain to convert it into an infrared spectrum. In a direct comparison of 5(a) and 5(b), the difference between the time-domain arrangement of the laser pulses and the interference pattern can be clearly seen.

[0129] Example 2

[0130] In the present invention, directly using Figure 5 (a) to calculate the singular spectrum, and subsequent data mining gives the molecular concentration of the test sample that constructs the absorbance signal, such as Figure 6 As shown. The singular spectrum represented by multifractal dimension and singular intensity is fundamentally different from the ordinary infrared spectrum represented by light intensity and wave number diagram in many aspects. First, the key parameter in the molecular absorption spectrum is the intensity of the spectral peak, so it is read from the y-axis; in contrast, the key parameters in the singular spectrum are the three primary key singular intensities, so they are read from the x-axis. Second, the molecular concentration is mined from the measured data of the spectral peak. Since the molecular absorption spectrum may have many spectral peaks, the molecular concentration result obtained depends on the artificial identification and selection of the key spectral peaks in the data processing. There is an inevitable uncertainty in the obtained molecular concentration. In contrast, a set of absorbance signals always produces the same singular spectrum and a set of unique primary key singular intensities, from which the molecular concentration can be data mined with the singular intensities and singular bases. Therefore, the uncertainty of the molecular concentration obtained by the key singular coefficient analysis of the present invention is extremely small. Third, in traditional spectral analysis, the absorbance of the spectral peak is usually linearly related to the molecular concentration; therefore, a linear regression model is usually used in traditional spectroscopy, and the unknown molecular concentration is calculated based on the measured absorbance with reference to the linear regression model. However, in one illustrative embodiment of the present invention, the key singular coefficients recorded as a function of the molecular concentration are as follows: Fig. 9 As shown, it is definitely not a simple linear correlation, but a complex correlation. Therefore, the present invention adopts the most advanced method of building and training machine learning regression models to mine molecular concentrations from multiple key singular coefficient data.

[0131] Example 3

[0132] In order to evaluate the prediction accuracy of the method and device disclosed in the present invention, patients with 41 different blood glucose concentrations were used as test samples, and the method and device disclosed in the present invention were used to predict the blood glucose concentration of the patient. Figure 4 Blood glucose was measured using the specific device configuration shown in FIG. These specific test samples were also tested using a standard blood glucose meter at nearly the same time and under the same conditions. Fig.10 In the standard Clarke error grid diagram shown in (a), the blood glucose concentrations obtained using the method and apparatus of the present invention (concentrations obtained by the new apparatus and new method) have an excellent correlation with the blood glucose concentrations obtained using a standard blood glucose meter. Obviously, the correlation between the concentrations measured by the new instrument and new method and the concentrations measured by the blood glucose meter is very good. More specifically, all 41 concentrations measured by the new instrument and new method are within the area within the grid where the error does not exceed 15% compared to the concentration measured by the blood glucose meter. This example demonstrates the applicability and impact of the method and apparatus of the present invention.

[0133] Example 4

[0134] The same set of time-domain absorbance signals in Example 3 was used with conventional spectroscopy to determine the molecular concentration. Specifically, the same set of time-domain absorbance signals was segmented into a segment whose duration corresponds to the span of wavenumber variation covered by the QCL used in the device described in Example 3. The precise range of wavenumber variation is 193 cm -1 , between 935 and 1128 cm -1 There are 72 segments with the same span, each with about 28,000 recorded data points, which are correctly collapsed, arranged and tiled into a single combined segment, which is similar to the infrared molecular absorption spectrum of blood glucose. Figure 8 41 infrared molecular absorption spectra were plotted, which were transformed from 41 sets of time-domain absorbance signals. Figure 8 Also included is an infrared molecular absorption spectrum recorded using a state-of-the-art FTIR spectrometer, a Nicolet iS50 FTIR spectrometer, by pressing the hypothenar of the patient at the time of the highest concentration of glucose in the same patient against the attenuated total reflectance window. Obviously, the shape of the infrared molecular absorption spectrum transformed from the time-domain absorbance signal collected by the device of the present invention matches well with the shape of the infrared molecular absorption spectrum recorded by the Nicolet iS50 FTIR spectrometer. The 41 infrared molecular absorption spectra converted from the time-domain absorbance signal are then used to calculate the concentration from the spectral data using standard spectral methods. Since the prior art has proven that integrating machine learning data mining methods with conventional spectral molecular concentration analysis can effectively improve the accuracy of spectral molecular concentration analysis, this method is used in this example. Specifically, the infrared molecular absorption spectra are first subjected to a standard principal component analysis, and then data mining is performed using a support vector machine regression model to obtain 41 molecular concentrations (referred to as the concentrations obtained by the new instrument and conventional spectral analysis methods). These concentrations obtained by the new instrument and conventional spectral analysis methods are then paired with the reference data of the blood glucose meter in the Clarke error grid diagram, such as Fig.10 (b) Clearly, the correlation between the concentrations measured by the new instrument and conventional spectroscopy and the concentrations measured by the blood glucose meter is only marginally acceptable and far inferior to that of the blood glucose meter. Fig.10 (a) shows the correlation of the concentrations measured by the new instrument and the new method. More specifically, of the 41 concentrations, only 78% of the concentrations were within the region of the grid where the error compared to the concentration measured by the blood glucose meter was no more than 15%. This example shows that even if machine learning techniques are applied to improve this conventional spectral method, the method and apparatus of the present invention are significantly better than the results obtained by replacing the new method of the present invention with the conventional spectral method.

[0135] Example 5

[0136] In Examples 3 and 4, by using the Figure 3 and 4 The device described in the above structure is used to measure the absorbance signal in time series. Fig.10 Illustrative blood glucose concentrations shown in (a) of the new apparatus and method with superior accuracy performance to a reference blood glucose meter. In this example, in order to further reduce the size and weight of the device of the present invention, and to measure the practicality of integrating all hardware components onto a single printed circuit board, Figure 3 and 4 The hardware design of the specific device described in Example 1 was re-examined. In this example, for the sake of clarity, the demonstration Fig.10 (a) shows the excellent accuracy performance of the hardware ( Figure 4 ) is called Hardware Design #1. The first part of the study reviews the conclusions of using an ADC oscilloscope in Hardware Design #1. This ADC has a very high sampling speed of 100MHz. Fig.10 In the precise experiment shown in (a) that produced excellent precision performance, an ADC rated at 100 MHz was chosen because the duration of each absorbance signal pulse reaching the infrared detector was 100 ns, and the purpose of the experimental design was to sample this pulse quickly enough to obtain an accurate pulse shape with 10 data points. However, the cost of this design is considerable. First, the entire cycle of a "scan" for molecular analysis lasts about 20 ms, and a typical measurement of a molecular concentration includes 3-10 scans. For each complete cycle, the ADC of hardware design #1 outputs 2 million data points, and for a complete measurement, it outputs up to 20 million data points. Therefore, this hardware component gives a large amount of data at a high speed, which is very expensive. Physically, it has dimensions of 15cm*12cm*2.5cm and weighs 0.5kg. In this case study, this ADC was replaced by an ADC rated at 20MHz, which has physical dimensions of 5cm*2.5cm*1cm and weighs 10g, which is 50 times lighter than the original ADC. After replacing the ADC, the concentrations measured by the new instrument and new method were compared with the concentrations of the reference blood glucose meter, e.g. Fig.10 (c) As shown. Clearly, the accuracy performance is still very good. Specifically, more than 97% of the measurements are confined within the error grid area with an error of no more than 15%. It should be noted that the FDA recommends an accuracy threshold of 95% for non-invasive blood glucose analyzers with an error of no more than 15% to replace conventional blood glucose meters.

[0137] Following the successful optimization of Hardware Design #1, other hardware modifications were investigated. A new QCL system was considered to replace the QCL system in Hardware Design #1. Specifically, the size of the QCL system in Hardware Design #1 is

[0138] 11.2cm*8.2cm*8.8cm, weight is 1.20kg. Its operating frequency is 180kHz, bandwidth is 193cm -1 , with a rated power of over 600mW. A model that can replace it has a physical size of 4mm*4mm*2mm and weighs less than 10g. In addition, in the updated version of hardware design #1, a programmable logic controller is used as the new MCU. In addition, the light-sample interaction unit 200 of hardware design #1 is a multiple attenuated total reflection prism (MATR) with a physical size of 12cm*10cm*9cm and a weight of 0.9kg. The improved version is much smaller and lighter.

[0139] Overall, the total physical size of Hardware Design #1 is more than 10 times smaller than the state-of-the-art FTIR spectrometer, and the weight is reduced from 70kg to 5kg. The new version of Hardware Design #1 is 10-100 times smaller and 10-100 times lighter than the original. The miniaturized upgraded laser system 100, the simplified and miniaturized MATR 200, the simplified and miniaturized detector and ADC components 300, and the miniaturized upgraded MCU 400 are integrated on a ceramic printed circuit board using surface mount technology, which has a better hardware miniaturization effect. The best hardware design is to integrate all these components into a single chip.

[0140] The description and examples of the present invention provide a complete description of the structure and use of the illustrative embodiments. Although certain embodiments have been described with a certain degree of particularity or with reference to one or more separate embodiments, those skilled in the art may make many modifications to the disclosed embodiments without departing from the scope of the present invention. Therefore, it is not intended that the various illustrative embodiments of the device be limited to the specific forms disclosed. On the contrary, it includes all modifications and alternatives that fall within the scope of the claims, and other embodiments outside the illustrated embodiments may include some or all of the features of the illustrated embodiments. For example, components may be omitted or combined into a single structure, and / or may be connected interchangeably. In addition, where appropriate, aspects of any embodiment described above may be combined with aspects of any other embodiment described to form further embodiments having comparable or different properties and dealing with the same or different problems. Similarly, it should be understood that the benefits and advantages described above may relate to one embodiment or may relate to several embodiments.

[0141] Furthermore, unless the phrases "means for" or "step for" are explicitly used in a particular claim, these claims are not intended to include or should be interpreted as including functional or step limitations.

[0142] The particular embodiments disclosed above are illustrative only, as the application may be modified and implemented in different but equivalent manners which will be apparent to those skilled in the art having the benefit of the teachings herein. It is therefore apparent that the particular embodiments disclosed above may be altered or modified, and all such variations are contemplated as being within the scope and spirit of the application. The protection sought herein is therefore as described in the specification. It is apparent that an application having significant advantages has been described and illustrated. While the present application is shown in a limited number of forms, it is not limited thereto, but is susceptible to various changes and modifications without departing from the spirit thereof.< / x> < / x>

Claims

1. A method for accurately measuring molecular concentration using singular coefficient analysis technology, characterized in that: The method comprises: directly mining the time-domain absorbance signal of the test sample to obtain a group of key singular coefficients, and then mining the molecular concentration of the test sample from a plurality of key singular coefficients.

2. The method according to claim 1, characterized in that The time-domain absorbance signal contains the fingerprint absorbance characteristics of the molecular components in the gas, liquid or solid test sample, and the time-domain absorbance signal changes continuously in a time series according to the change of the wave number of the monochromatic light over time.

3. The method according to claim 1, characterized in that The process of directly data mining the time-domain absorbance signal of the test sample to obtain a set of key singular coefficients includes four consecutive steps of multifractal detrended fluctuation analysis or similar singular coefficient analysis, and the four consecutive steps include the step of calculating a filtered absorbance signal profile from the time-domain absorbance signal, the step of calculating a variance function from the filtered absorbance signal profile, the step of calculating a series of q-order fluctuation functions from the variance function, and the step of calculating singular cardinalities and singular spectra from the series of q-order fluctuation functions to construct a set of key singular coefficients.

4. The method according to claim 3, characterized in that The steps of calculating a filtered absorbance signal profile from the time-domain absorbance signal, calculating a variance function from the filtered absorbance signal profile, calculating a series of q-order wave functions from the variance function, and calculating singular bases and singular spectra from the series of q-order wave functions to construct a set of key singular coefficients include taking the absorbance x of the i-th time unit in the time-domain absorbance signal i , calculate its average value when the total duration is N time units <x> , and construct the filtered absorbance signal profile Y(i):< / x> The signal Y(i) is divided into M non-overlapping segments of equal time length s from i=1 to i=N, and then divided into another group of M non-overlapping segments of time length s from i=N to i=1. Then the local trend y of the vth segment is calculated by the least squares fitting of the polynomial function. v (i) and determine the variance function Y of the vth segment var (v,s): From Y var (v,s) calculates a series of q-order wave functions F(q,s): The singular cardinality β, singular intensity α, and multifractal dimension f(α) are calculated from F(q,s) to construct the singular spectrum f(α), where h(q) is calculated by d(log(F)) / d(log(s), and h′(q) is calculated by d(h) / d(q) α=h(q)+qh′(q)# f(α)=q 2 h′(q)+1#。 5. The method according to claim 1, characterized in that The constructing of a set of key singular coefficients includes calculating a singular cardinality and extracting three first-order key singular intensities consisting of the singular intensities of the leftmost point, the rightmost point and the maximum point from the singular spectrum, and deriving a plurality of second-order key singular intensities from the first-order key singular intensities; The plurality of secondary key singular intensities include a secondary key singular intensity defined as a width between a leftmost point and a rightmost point.

6. The method according to claim 1, characterized in that The data mining of the molecular concentration of the test sample from multiple key singular coefficients includes: training a machine learning regression model using multiple key singular coefficients of several reference samples with known molecular concentrations, and determining the unknown molecular concentration of the test sample by projecting the key singular coefficients of the test sample onto the trained machine learning regression model; the reference sample is a sample with the same properties as the test sample, except that the molecular concentration of the reference sample is known; the machine learning regression model includes a standard support vector machine regression model, or any other machine learning regression model.

7. A device for accurately measuring molecular concentration using singular coefficient analysis technology, characterized in that: The device includes a main control unit and a laser system, a light-sample interaction unit, a detector, and an analog-to-digital converter connected thereto; the laser system is used to generate the required laser pulses, the light-sample interaction unit is used to complete the reaction between the laser pulses and the test sample, the detector is used to detect the reaction results in the light-sample interaction unit, and the analog-to-digital converter is used to convert the results of the detector into analog-to-digital and output them to the main control unit. The main control unit includes hardware programmed with software to act as a singular coefficient calculator and a machine learning regression model processor, and is used to complete the precise measurement of molecular concentration according to the output results of the analog-to-digital converter.

8. The device according to claim 7, characterized in that The main control unit includes an integrated circuit chip, or multiple integrated circuit chips packaged using surface mount technology on a printed circuit board, or a programmable logic controller, or a microcomputer or a computer, including a telecommunication device that can transmit molecular concentration information to a visualization mobile terminal.

9. The device according to claim 7, characterized in that The software-programmed hardware of the main control unit includes a singular coefficient calculator for data mining singular spectra from the absorbance signal output by the analog-to-digital converter in the time domain and constructing a set of key singular coefficients; and a machine learning regression model processor for data mining molecular concentrations from a set of key singular coefficients output by the singular coefficient calculator; the machine learning regression model processor includes software for setting up and applying a support vector machine regression model to data mine molecular concentrations from a set of key singular coefficients output by the singular coefficient calculator.

10. The device according to claim 7, characterized in that The laser system includes a broadband laser and a spectral filter for emitting time-domain monochromatic pulses, and the wave number of the pulses arranged in the time domain also changes sequentially with time; wherein the broadband laser is a quantum cascade laser, and the quantum cascade laser includes a direct or indirect connection to the main control unit to receive and track the control signal from the main control unit, and then emits laser pulses with a frequency of 10kHz to 1MHz, 100kHz to 500kHz or 150kHz to 200kHz in the time domain; the broadband laser includes a wavelength from 20 to 300cm -1 , from 40 to 200cm -1 or 200cm -1 The bandwidth of the spectral filter is 0.05-4 cm; the spectral filter includes a grating whose rotation and filtering wave number are manipulated by a micro-electromechanical actuator or a grating whose length and filtering wave number are controlled by voltages applied to two end electrodes of the grating plate. The spectral resolution of the spectral filter is controlled by a plurality of apertures along the optical path of the laser to achieve filtering of the laser pulse. The spectral resolution is set to 0.05-4 cm -1 、1-4cm -1 or 1.6cm -1 ; The filtered laser pulse width was set to 50-500ns, 50-200ns, 50-100ns or 100ns.

11. The device according to claim 7, characterized in that The light-sample interaction unit comprises an infrared transparent unit for accommodating a test sample in a gas, liquid or solid state, and the infrared transparent unit comprises an infrared transparent single attenuation total reflection prism or a multiple attenuation total reflection prism.

12. The device according to claim 7, characterized in that The detector comprises an InAsSb infrared detector or an MCT infrared detector.

13. The device according to claim 7, characterized in that The analog-to-digital converter includes an analog-to-digital converter operating at a sampling frequency of 10-100 MHz or 20-60 MHz, including an n-bit analog-to-digital converter with n ranging from 10 to 24.

14. The device according to claim 7, characterized in that The laser system, light-sample interaction unit, detector, analog-to-digital converter and main control unit are packaged on a plastic printed circuit board or a ceramic printed circuit board using surface mounting technology, or are highly integrated into a single-chip laser molecular analyzer using chip integration technology.

15. The device according to claim 7, characterized in that The device is designed as a portable molecular analyzer, which is more than 10 times smaller and more than 10 times lighter than a typical FTIR spectrometer; or the device is designed as a wearable molecular analyzer, which is more than 100 times smaller and more than 100 times lighter than a typical FTIR spectrometer.

Images