Absorption spectrum tomography system based on channel multiplexing and data real-time compression

Abstract

The invention provides an absorption spectrum tomography system based on channel multiplexing and data real-time compression. The system comprises a laser generation module, a tomography sensor, a data acquisition module, a channel multiplexing module, a data compression processing module, a data transmission module, an upper computer and the like. Performing multi-angle projection time-sharing scanning in the tomography sensor; grouping the detectors in the sensor, wherein each group comprises two detectors; each group of detectors are connected to the same acquisition channel of the data acquisition module by utilizing the channel multiplexing module, and the optimal duty ratio of effective absorption spectrum signals is realized by combining a laser starting time sequence in the laser generation module; the data compression processing module utilizes on-chip wavelet basis decomposition operation to realize real-time compression and storage of sampling data with a variable compression rate. Under the condition that the original characteristics of the data are not changed, the number of acquisition channels is reduced by nearly half; meanwhile, the signal-to-noise ratio of absorption spectrum data is increased, the sampling time is greatly prolonged, and the method has a wide application prospect.

Description

Technical Field

[0001] The invention relates to an absorption spectrum tomography system, in particular to an absorption spectrum tomography system based on channel multiplexing and real-time data compression, and belongs to the technical field of laser absorption spectrum and gas parameter measurement. Background Art

[0002] Combustion diagnostic technology aims to monitor and analyze key parameters in the combustion process in order to optimize and adjust the combustion process in real time. The combustion process will affect the main parameters such as the air temperature around the flame and the concentration of gas components. Therefore, measuring the gas temperature and the concentration of the main components around the flame is an important means of monitoring the combustion process. In recent years, with the emergence and continuous development of Tunable Diode Laser Absorption Spectroscopy (TDLAS) technology, its wide measurement range, high sensitivity and non-invasive characteristics have demonstrated its applicability in combustion measurement.

[0003] TDLAS flow field diagnosis is divided into one-dimensional diagnosis and two-dimensional diagnosis in terms of measurement dimension. One-dimensional diagnosis is based on the measurement of line of sight average, measuring the weighted average of flow field parameters (temperature, concentration, etc.) on the optical path. In 2016, Zhanrong Zhang et al. published a paper titled "Detection of gas temperature using a distributed feed-back laser at O2 absorption wavelength 760nm" in the Journal of Optical Technology, Vol. 83, p. 673. They measured the gas temperature in a tube furnace in the temperature range of 300-900K using the direct absorption method. The temperature measurement accuracy at low temperatures is better than that at high temperatures. In 2022, Xiaonan Liu et al. published the paper "Tunable Diode Laser Absorption Spectroscopy Based Temperature Measurement with a Single Diode Laser Near 1.4μm" in Sensors, Vol. 22, No. 16, No. 6095. They used a distributed feedback diode whose emission wavelength simultaneously covered two water absorption lines to measure the temperature of the combustion field, calibrated the output wavelength of the diode laser, and used the system to perform temperature tests on the scramjet engine, proving that the system has good dynamic range and fast response characteristics. The one-dimensional diagnostic measurement method has a simple device structure and a simple and direct data processing algorithm, but it can only reflect the line-of-sight average value of the flow field state. Traditional TDLAS single-path measurement can only obtain the average value of the parameter to be measured on the absorption path. The measurement result lacks spatial resolution and cannot reflect the spatial distribution of the parameter to be measured. It cannot meet the diagnostic needs of complex combustion fields with uneven parameter distribution.

[0004] Therefore, with the advancement of tomography technology, scanning systems and image processing technology, two-dimensional diagnostic technology has gradually developed. Combining TDLAS technology with computer tomography technology to obtain laser absorption spectrum information at different angles and paths in the region of interest, and combining tomography algorithms to achieve two-dimensional distribution reconstruction of the parameters to be measured (such as temperature, gas component concentration, pressure, etc.) in non-uniform and complex combustion fields has gradually become a hot spot in combustion diagnosis. Zhenzhen Wang et al. published "Application of 2Dtemperature measurement to a coal-fired furnace using CT-TDLAS" in "Measurement Science and Technology" Vol. 31, No. 3, p. 035203 in 2020. A 32-channel two-dimensional CT-TDLAS system with circular and rectangular structures was proposed, which improves the accuracy and stability of the measurement and is suitable for the online two-dimensional temperature measurement of solid fuel reactions. In the article "Time resolved 2D concentration and temperature measurement using CT tunable laser absorption spectroscopy" published in 2015 in Flow Measurement and Instrumentation, Vol. 46, pp. 312-318, Y. Deguchi et al. used the absorption spectra of molecules such as water, ammonia and methane to design a 16-channel laser beam tomography system to measure the two-dimensional temperature, ammonia and methane distribution of engine exhaust and oscillating flames. The system has the potential of kHz response time and can perform real-time two-dimensional temperature and material concentration measurements in various industrial processes including engine applications. The TDLAS tomography system is convenient for projection from multiple laser paths and reconstruction of the two-dimensional distribution of temperature and gas species concentration. Since sufficient absorption spectrum data along different laser paths is required for reconstruction, the tomography data acquisition system needs to have a large number of channels and a large storage capacity, and the system also needs to have a high response speed and monitoring time during dynamic monitoring.

[0005] At present, in the measurement of absorption spectra, the commonly used commercial data acquisition systems are highly versatile, but have the disadvantages of being bulky, expensive, and unable to meet personalized needs. For this reason, more and more self-developed data acquisition systems are introduced into TDLAS tomography, especially when multiple laser signals are detected simultaneously at high sampling rates. The self-developed acquisition system can be customized according to functional requirements, highly personalized, and has a greater guarantee for data security. In the paper "Towards Parallel, 192 Channel, 40 MS / s / ch Data Acquisition for Optical Tomography: A System for Aero-Engine Exhaust Emission Diagnostics" published by E. Fisher et al. at the IEEE Sensor Conference in 2017, a parallel, high-speed, distributed data acquisition system was proposed, which adopts in-situ demodulation of tunable diode laser absorption spectrum signals, uses photodiode pre-amplification, filtering, digitization, signal demodulation, Ethernet encapsulation, and realizes microprocessor control on a multi-node, distributed basis. Mingxiang Li et al. designed an online TDLAS tomography system for measuring temperature and gas mole fraction distribution in the article "TDLAS Tomography System for Online Imaging and Dynamic Process Playback of Temperature and Gas Mole Fraction" published in IEEE Transactions on Instrumentation and Measurement, Vol. 71, No. 2006210 in 2022. The on-chip parameter extraction module based on the digital lock-in amplifier extracts absorption spectrum parameters at a speed of 1000 frames per second (FPS), and the real-time reconstruction frame rate on the host computer can reach 302 frames per second. The above system can detect multiple laser signals at a high sampling rate and realize imaging of high frame rate parameter distribution, but it needs to match more acquisition channels, which makes the hardware resources too much. In addition, the preservation of raw data is limited by the storage capacity of the storage chip, and it is impossible to obtain raw data of dynamic processes for a long time.

[0006] Increasing the duration of continuous measurement can be achieved by updating the data storage hardware or compressing the absorption spectrum data. Updating the data storage hardware is costly and involves upgrading the entire system, while compressing the absorption spectrum data can continue to increase the data storage capacity based on any storage hardware, which can solve the problem of insufficient storage capacity and thus increase the acquisition time of stored data.

[0007] Wavelet analysis was developed in the mid-1980s. It has the characteristics of multi-resolution, low entropy, and decorrelation. It is widely used in digital signal processing. Compared with traditional denoising algorithms, it has obvious advantages in analyzing and processing non-stationary signals such as absorption spectrum signals. Wavelet transform is a local transform of frequency and space (time). It is a breakthrough in Fourier transform. It can perform multi-scale refinement analysis on signals or functions through operations such as translation and scaling, and effectively extract important information from signals. In addition, in most practical cases, only noise samples of the original signal are available, and better performance can be obtained by using sub-band processing. Therefore, wavelet transform can provide a suitable model for denoising the absorption spectrum signal of the TDLAS system. Jingsong Li et al. proposed a new method based on discrete wavelet transform in "Wavelet Transform Based on the Optimal Wavelet Pairs for Tunable Diode Laser Absorption Spectroscopy Signal Processing" published in Applied Spectroscopy, Vol. 69, No. 4, pp. 496-506 in 2015. They selected the optimal wavelet pairs to adaptively process the tunable diode laser absorption spectrum for quantitative analysis, and verified that the wavelet transform can suppress noise, maintain peak shape and remove baseline. Hua Xia et al. described a wavelet transform signal processing strategy for gas monitoring in "Signal analytical processing based on wavelet transform fortunablediode laser absorption spectroscopy" published in Advanced Sensor Systems and Applications IV, volume 7853, article 785311 in 2010. According to the characteristics of TDLAS second harmonic signals, a signal enhancement method was designed using bior3.9 wavelet basis functions and multi-resolution decomposition.

[0008] Multiplexing technology refers to a method of integrating multiple channels on a transmission path, and then restoring the original mechanism or releasing each terminal channel. It is often used in data communications. Common multiplexing technologies include time division multiplexing technology, frequency division multiplexing technology, and channel multiplexing technology. Some multiplexing technologies are also used in TDLAS measurement systems. Fei Li et al. published "Simultaneous measurements of multiple flow parameters for scramjet characterization using tunable diode-lasersensors" in Applied Optics, Vol. 50, No. 36, pp. 6697-6707 in 2011. They coupled the laser beams of two bands of lasers and used time division multiplexing technology to separate the absorption signals. They used multiple measurement channels to simultaneously measure multiple important flow parameters of the scramjet combustion chamber. Ang Huang et al. introduced frequency-division multiplexing and main peak scanning methods into the wavelength modulation spectroscopy method in "Frequency-Division Multiplexing and Main Peak Scanning WMS Method for TDLAS Tomography in Flame Monitoring" published in IEEE Transactions on Instrumentation and Measurement, Vol. 69, No. 11, pp. 9087–9096 in 2020. For the first time, two wavelengths modulated with different frequencies were used to extract absorption in parallel along different laser paths to reconstruct the distribution of temperature and substance concentration at a frame rate of up to 10 kHz. Yiwen Feng et al. combined time division multiplexing technology with frequency division multiplexing technology in "Application of TDM and FDM methods in TDLAS based multi-gas detection" published in "Optical And Quantum Electronics" Volume 53, Article 195 in 2021. They combined the advantages of both technologies to achieve the effect of using minimal equipment and saving time, thus realizing the measurement of trace gases.Fanghao Lu et al. introduced wavelength division multiplexing technology in "Beam-steering-effect immune laser absorption spectrum extraction for precise and robust temperature measurement" in "Measurement" Vol. 231, p. 114549 in 2024 to detect the distorted signal and the distorted information separately, adaptively filter the distorted signal, and use it to proportionally compensate for the beam steering distortion parasitic on the absorbed laser signal. Since the diagnostic technology requires more detectors to ensure the resolution of the reconstruction when reconstructing two-dimensional or three-dimensional distribution, the acquisition system needs to have more channels to collect the output signal of the detector.

[0009] In summary, in order to realize the rapid analysis of combustion diagnostic data and long-term monitoring of the combustion field, it is necessary to solve the problems of large data volume, limited sustainable measurement time, and many acquisition channels in the tomography system. To this end, the present invention proposes an absorption spectrum tomography system based on channel multiplexing and real-time data compression. In view of the problem of many acquisition channels in the tomography system, a channel multiplexing module based on the time-sharing scanning method of the multi-angle tomography sensor is designed. Combined with the laser start-up timing in the sensor, the two detector outputs are multiplexed into one acquisition input to achieve the optimal duty cycle of the effective absorption spectrum signal and improve resource utilization; in view of the problem of large data volume and limited duration in the tomography system, an on-chip data compression module is designed to achieve the effect of not reducing the sampling rate and response speed when the storage chip is fixed, so that the amount of stored data and the acquisition time are greatly increased, and the signal-to-noise ratio of the absorption spectrum data is improved, which is convenient for signal post-processing, and has broad application prospects in the field of laser absorption spectrum gas parameter measurement. Summary of the invention

[0010] 1. Technical issues to be resolved

[0011] The purpose of the present invention is to provide an absorption spectrum tomography system based on channel multiplexing and real-time data compression, which is used to solve the problems of multiple sampling channels and long acquisition time required for the acquisition system in the tomography system. A channel multiplexing module based on the time-sharing scanning mode of a multi-angle sensor is designed, and combined with the laser start-up timing in the tomography system, the optimal duty cycle of the effective absorption spectrum signal is achieved, thereby improving resource utilization; and an on-chip data compression module is designed, which achieves the effect of not reducing the sampling rate and response speed when the storage chip is fixed, thereby greatly increasing the amount of stored data and the acquisition time, and improving the signal-to-noise ratio of the absorption spectrum data, thereby facilitating signal post-processing.

[0012] (II) Technical solution

[0013] The present invention is an absorption spectrum tomography system based on channel multiplexing and real-time data compression, and its main implementation process includes the following steps:

[0014] Step 1: Based on the laser start-up timing in the sensor, design the detector grouping and connection method, multiplex the two detector outputs into one acquisition input, and maximize the duty cycle of the acquisition system.

[0015] The structure of the tomography sensor is designed as a regular N-gon, that is, it contains 2*N tunable diode lasers. The laser generation module controls the tunable diode lasers in the tomography sensor to output scanning lasers in a time-division multiplexing manner. The laser output by each laser is coupled into a beam of light using a fiber coupler. After being transmitted through the optical fiber, it is collimated by the collimator on the fan beam sensor and widened into a fan-shaped laser beam by the Powell prism. The beam expansion angle is θ°. The output laser passes through the gas to be measured in the area of ​​interest of the tomography sensor, and the transmitted laser is received by y photodetectors at the same time. The beam expansion angle θ° and the number of detectors y that receive valid signals at the same time can be determined according to the size of the area of ​​interest. If a detectors are installed on each side of the sensor, the total number of detectors Y contained in the tomography sensor of N angles is N*a, and y needs to satisfy the conditions of a≤y≤2a, and the left and right numbers are symmetrical, which can ensure that the laser signals at each angle do not interfere with each other when there is a large area of ​​interest. At this time, each detector will receive laser signals at two angles. If the scanning time of a single-angle laser signal is T r , the time for each detector to receive the signal is 2*T r , the time required to scan all N angles is N*T r , then the duty cycle of the detector output signal is

[0016] When Y is an even number, the detector is divided into Each group has two detectors and they do not receive valid signals at the same time. The two detectors in each group are connected to the same acquisition channel. That is, when laser 1 is working, the corresponding y detectors are in working state, and the other detectors are in non-working state. According to the number of sensor angles, two groups of detectors that do not interfere with each other when lasers at different angles are turned on can be divided. They are connected to the same group of acquisition channels in the scanning order to realize the multiplexing of the acquisition channels and improve the duty cycle of the system to

[0017] When Y is an odd number, the detector is divided into Group, front Each group has two detectors, and the two detectors in each group are connected to the same acquisition channel. The last group has only one detector, which is connected to one acquisition channel alone. The duty cycle of the system is still increased to

[0018] Step 2: Based on the selected wavelet basis function, decomposition layer number and threshold range as the initialization parameters of the data compression processing module, select the field programmable gate device FPGA as the processor, perform wavelet decomposition on the transmitted laser intensity signal to solve all detail coefficients and approximate coefficients in real time, perform threshold processing and quantization on the coefficients in the FPGA, and splice them into a signal form suitable for storage. Finally, the compressed signal is unspliced ​​and reconstructed by wavelet by the host computer to obtain the reconstructed signal. The detailed process is as follows:

[0019] The first step is to select parameters such as wavelet basis function, decomposition level and threshold range as initialization parameters of the data compression processing module.

[0020] The original signal is decomposed into a single layer using the lifting wavelet transform method to obtain approximate coefficients and detail coefficients. The wavelet transform method is:

[0021]

[0022] in, and ψ j,k (t) are called scaling function and wavelet function respectively, A j,k and D j,k are the approximate coefficients and detail coefficients. The multi-resolution analysis algorithm is introduced, the decomposition layer number j is preset, the detail coefficients of each layer are thresholded, and new detail coefficients are obtained. The approximate coefficients of the last layer are retained, and each coefficient is reconstructed by wavelet. The number of decomposition layers is changed, the signal-to-noise ratio of the reconstructed signal under different decomposition coefficients is calculated, and the number of layers with the smallest signal-to-noise ratio is selected to obtain the denoised absorption spectrum signal, and the wavelet basis function is changed to repeat the above operation, and the reconstruction results are compared to select the appropriate solution. The three parameters of wavelet basis function, decomposition layer number and threshold range are retained as the initialization parameters of the data compression processing module.

[0023] In the second step, the FPGA processor in the data compression processing module is used to perform on-chip wavelet decomposition on the sampled signal, calculate all detail coefficients and approximate coefficients in real time, and perform threshold processing and quantization processing on the required coefficients in the FPGA.

[0024] The parallel data acquisition module quantizes the channel multiplexed transmitted laser intensity signal, and the FPGA processor controls it to send the digital signal of the laser intensity to the data compression processing module in real time to realize the on-chip wavelet transform. The FPGA performs streaming parallel processing on the input data, that is, it does not need to cache the data or intermediate results in the external dynamic random access memory DRAM for buffering, but only uses the on-chip memory resources. This streaming design ensures high throughput and low latency of the processing process. In the data compression processing module, the signal is calculated in binary data mode, and the data selection control signal and multiple registers are set. The selection signal is used to control the sequence selection of the register to select the subsequent calculation order and the output data order to complete the prediction and update steps of the lifting wavelet transform. All the detail coefficients and approximate coefficients of the multi-layer wavelet transform are obtained in sequence, and the on-chip implementation of the lifting wavelet transform is realized. This method has a relatively simple hardware design, saves resources, and is flexible to use. Only the multi-layer detail coefficients and the last layer of approximate coefficients of the demodulated original absorption spectrum signal are retained, and the number of bits of the retained coefficients is quantized, and the number of bits of the coefficients is reduced without affecting the range of the coefficient value, so as to realize the data compression function.

[0025] The third step is to perform single-channel data splicing operation on the compressed coefficients.

[0026] Since the multi-layer wavelet transform hardware implementation process needs to continuously perform frequency division operations, each time a layer of wavelet transform is performed, the main clock frequency of the wavelet transform of this layer is half of the main clock frequency of the wavelet transform of the previous layer. Therefore, splicing processing is required before the quantized coefficients are stored in the subsequent storage; first, after sampling by the analog-to-digital converter, an absorption spectrum signal with a point number of n and a clock frequency of v is obtained. The frequency of the detail coefficients and approximate coefficients obtained after the first layer of wavelet transform is v1, and the number of points is The frequency of the detail coefficients and approximate coefficients obtained after the second layer of wavelet transform is v2, and the number of points is The frequency of the detail coefficients and approximate coefficients obtained after the j-th layer wavelet transform is v j , the points are Then the frequencies after decomposition of each layer satisfy the following relationship:

[0027] v=2*v1=2 2 *v2=···=2 n *v j (2)

[0028] Then, the number of bits of the detail coefficients decomposed at the first layer is quantized to x1, the number of bits of the detail coefficients decomposed at the second layer is quantized to x2, and the number of bits of the detail coefficients decomposed at the jth layer is quantized to x j, the number of bits of the approximate coefficients obtained by the last layer of decomposition remains unchanged and is consistent with the number of bits of the original data, which is x; finally, the clock domain is unified by beating, and the coefficients obtained after compression are aligned, with the period of the coefficient with the smallest frequency as the alignment unit. Then the spliced ​​binary result l1 of the single-channel acquisition of n absorption spectrum data after compression is:

[0029]

[0030] The fourth step is to realize data stitching of acquisition modules containing multiple acquisition channels.

[0031] One acquisition module is designed with four acquisition channels, so the data compression module also needs to ensure the simultaneous compression of four channels of data. After each channel of data enters the FIFO buffer, it is input into the wavelet transform data compression module for j-layer wavelet decomposition. The splicing results of the four channels are obtained according to the number of quantization bits. The quantization and splicing methods of each channel are consistent. It is necessary to add a two-bit flag prefix to the splicing result of each channel. Channel 1 is 00, channel 2 is 01, channel 3 is 10, and channel 4 is 11. The results after adding the flag are l1′, l2′, l3′ and l4′. The output results are independent of each other and the clock domain is consistent. The splicing result of the four channels is the first bit of each channel after adding the flag:

[0032] l=[l1',l2',l3',l4'] (4)

[0033] Step 5: Calculate the compression ratio CR of the compression module by the ratio of the data size before and after signal compression:

[0034]

[0035] According to the above splicing results, the actual compression ratio of the data compression module can be calculated as:

[0036]

[0037] The actual compression ratio depends on the set quantization bit number and the number of decomposition layers. The smaller the quantization bit number and the higher the number of decomposition layers, the greater the compression ratio. j When =0, the compression ratio reaches the maximum value.

[0038] Step 6: Convert the spliced ​​compression result into a data format suitable for writing into the storage chip. After caching in the storage chip, the data transmission module uploads all the data to the host computer to de-splice the compressed signal and reconstruct it by wavelet to obtain the reconstructed signal.

[0039] Beneficial Effects

[0040] The beneficial effect of the present invention is that an absorption spectrum tomography system based on channel multiplexing and real-time data compression is proposed. A channel multiplexing module based on the time-sharing scanning mode of the tomography sensor is designed to group the detectors working in different time periods, and each group is connected to an acquisition channel. The hardware implementation is simple, and the optimal duty cycle of the effective absorption spectrum signal is achieved, thereby improving the utilization rate of hardware resources. At the same time, the design of the data compression processing module can realize on-chip real-time compression of the absorption spectrum data. According to the initialization parameters of the custom wavelet transform, the wavelet decomposition is implemented on the FPGA, and the real-time compression and storage of the sampled data with a variable compression rate are realized. The system can solve the problems of large amount of collected data, many collection channels, and long sampling time in the TDLAS system. While the original characteristics of the data remain unchanged, the number of collection channels is reduced by nearly half; at the same time, the signal-to-noise ratio of the absorption spectrum data is improved, and the sampling time is greatly increased, providing a new idea for the optimization design of the laser absorption spectrum tomography system. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Attached Figure 1 :Flowchart of an absorption spectral tomography system based on channel multiplexing and real-time data compression

[0042] Attached Figure 2 :Schematic diagram of an absorption spectral tomography system based on channel multiplexing and real-time data compression Attached Figure 3 : Schematic diagram of the five-angle tomography sensor

[0043] Attached Figure 4 :The signal-to-noise ratio results of the reconstructed signal by lifting wavelet transform with different decomposition levels DETAILED DESCRIPTION

[0044] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is further described in detail below in combination with specific embodiments and with reference to the accompanying drawings.

[0045] Instructions attached Figure 1 is a flow chart of an absorption spectrum tomography system based on channel multiplexing and real-time data compression according to an embodiment of the present invention; Figure 2 1 is a schematic diagram of the structure of an absorption spectral tomography system based on channel multiplexing and real-time data compression according to an embodiment of the present invention, which mainly includes a laser generating module 101, a tomography sensor 102, a channel multiplexing module 103, a parallel data acquisition module 104, a data compression processing module 105, a data transmission module 106 and a host computer 107.

[0046] The laser control module 101 includes 5 output lasers with a central wavelength of v1 = 7185.6 cm -1 The laser and the 5 output lasers have a central wavelength of v2 = 7444.4 cm-1 There are 10 lasers in total. One laser has a central wavelength of v1 = 7185.6 cm -1 A laser with a central wavelength of v2 = 7444.4 cm -1 The lasers of the five-angle tomography sensor 102 are grouped as one, and the ten lasers are divided into five groups, corresponding to the five angles of the sensor. The two lasers in each group output scanning lasers by time division multiplexing, and are coupled into one light by a fiber coupler. After being transmitted through the fiber, they are collimated by the collimator on the five-angle tomography sensor 102, and then widened into a fan-shaped laser beam by a Powell prism. The detailed process is carried out according to the following steps:

[0047] Step 1: Design the grouping and connection mode of the detectors according to the channel multiplexing module 103 based on the time-sharing scanning mode of the tomographic imaging sensor 102. The embodiment of the present invention adopts a tomographic imaging sensor 102 with five angles, and 12 photodetectors are installed in parallel on each side of the five-angle tomographic imaging sensor 102. Two laser sources centered on different wavelengths are coupled into an optical fiber, and a fan-shaped beam is formed through each lens, and the beam passes through the flame in the center. The attenuated light signal is detected by a detector installed on a pentagonal structure. The laser from each angle is received by 24 detectors, that is, 12 detectors are directly opposite and 6 detectors are on each adjacent side. In each spectral scan, 10 lasers from 5 vertices cascade to emit spectral light, and the intensity is collected by the detector. The lasers at each angle adopt a time-sharing scanning mode. When each laser is turned on, the 24 photodetectors at the corresponding angle collect the light intensity signal. At this time, the output of the remaining detectors is 0. The status of each detector when each laser is turned on is listed, as shown in Table 1:

[0048] Table 1 Detectors receiving optical signals when the laser is turned on

[0049]

[0050] It can be seen from the information in the table that the corresponding detector serial numbers when any two lasers are turned on overlap, which means that in the set working state, only one laser can be turned on for each scan, and the other lasers are turned off. The five lasers are scanned to complete a scanning cycle, and 60 laser detection signals are obtained in one week. The parallel data acquisition module 104 is composed of multiple acquisition boards, each board is designed with 4 acquisition system inputs, and eight boards can measure 32 signal inputs at the same time. The cache size of each acquisition board is the same. Time-division multiplexing multi-channel signal detection is adopted. It is necessary to ensure that the number of detectors allocated to each board is the same, so that the data collected by each board is evenly distributed in time. The signal connection scheme of the multi-channel acquisition system is shown in Table 2:

[0051] Table 2 Design of signal connection between multi-channel acquisition board and detector

[0052]

[0053] Among them, each acquisition board uses 4 acquisition signals, and each acquisition signal input will be connected to the signals of two detectors. For example, acquisition board No. 1, its signal input channel 1 will be connected to detectors No. 7 and No. 37. This connection method uses 30 synchronous input channels instead of 60 synchronous input channels, which doubles the duty cycle of the acquisition system.

[0054] Step 2: Select wavelet basis function, decomposition layer number and threshold range as initialization parameters of data compression processing module 105, select field programmable gate circuit device FPGA as processor, perform wavelet decomposition on the transmitted laser intensity signal to calculate all detail coefficients and approximate coefficients in real time, perform threshold processing and quantization processing on the coefficients in FPGA, and splice them into a signal form suitable for storage. Finally, the compressed signal is de-spliced ​​and reconstructed by wavelet by the host computer to obtain the reconstructed signal. The detailed process is as follows:

[0055] The first step is to select parameters such as wavelet basis function, decomposition level number and threshold range as initialization parameters of the compression processing module 105.

[0056] The lifting wavelet method can solve the problems of computational complexity and the generation of a large number of floating-point numbers in ordinary discrete wavelet transforms, while possessing all the excellent characteristics of discrete wavelet transforms. Among them, the 5 / 3 lifting wavelet transform is one of the most commonly used lifting schemes. Its computational complexity is only half of that of discrete wavelet transforms, and its calculation speed is fast. It can realize in-situ calculations and integer lifting. The transformation results are interlaced and arranged, which is suitable for FPGA real-time compression processing of absorption spectrum signals. In noise reduction, the larger the number of wavelet decomposition layers, the more obvious the different characteristics of noise and signal performance, and the more conducive to signal-noise separation; but for reconstruction, the more decomposition layers there are, the greater the distortion, that is, the greater the reconstruction error. From the attached Figure 4 It can be seen that when the signal-to-noise ratio is set to 40-60dB, the lifting wavelet transform with different layers L has a certain denoising effect on the WMS signal. As the signal-to-noise ratio of the original signal increases, the denoising effect decreases. Among them, when the decomposition layers are 2 and 3, the denoising effect on low signal-to-noise ratio and high signal-to-noise ratio signals is the best. Therefore, the embodiment of the present invention selects the lifting wavelet transform as the basis function for compression and denoising, and the decomposition layer number is set to 3 layers. The threshold selects the general threshold λ, and the calculation formula is:

[0057]

[0058] Among them, σ is the noise variance, and N is the length of the signal. The detail coefficient of the first layer of wavelet transformation is usually used to estimate the noise variance. In the process of quantization threshold processing of actual absorption spectrum data, a large amount of data from previous actual experiments is selected to estimate the noise variance, and the quantization threshold parameter that best suits the absorption spectrum signal is selected. The size of the general threshold can also be reduced according to the required compression ratio and the actual compression effect.

[0059] In the second step, the FPGA processor in the data compression processing module 105 is used to perform on-chip wavelet decomposition on the sampled signal, and all detail coefficients and approximate coefficients are solved in real time, and the required coefficients are threshold processed and quantized in the FPGA.

[0060] The parallel data acquisition module 104 quantizes the channel multiplexed transmitted laser intensity signal, and the FPGA processor controls it to send the digital signal of the laser intensity to the data compression processing module in real time to realize on-chip wavelet transform. The ADC in the parallel data acquisition module 104 acquires the absorption spectrum signal with 14-bit accuracy and 40MSPS sampling rate, and transmits the digital signal to the FPGA through the parallel LVCMOS interface. The data compression processing module 105 performs on-chip wavelet transform on the multi-channel signals collected by the parallel data acquisition module 104 according to the initialization parameters to obtain the demodulated multi-layer detail coefficients and the last layer of approximate coefficients, and quantizes the number of bits of each coefficient, when the reserved bits of d1, d2 and d3 are all set to 6 bits or all zero.

[0061] The third step is to perform single-channel data splicing operation on the compressed coefficients.

[0062] After the signal is sampled by AD, the clock frequency is 40MHz. The frequency of the detail coefficients and approximate coefficients obtained after the first wavelet transform is 20MHz. The frequency of the detail coefficients and approximate coefficients obtained after the second wavelet transform is 10MHz. The frequency of the detail coefficients and approximate coefficients obtained after the third wavelet transform is 10MHz. Through the beat method, the clock domain is unified, and the coefficients obtained after compression are aligned. The period of the coefficient with the smallest frequency is used as the alignment unit to obtain the splicing result of a single channel of data. When the number of reserved bits of d1, d2, and d3 are all set to 6 bits, the single-channel splicing result l1 is 56 bits; when the number of reserved bits of d1, d2, and d3 are all zero, the single-channel splicing result l1 is 14 bits.

[0063] The fourth step is to realize data stitching of the acquisition module containing four acquisition channels.

[0064] A single parallel data acquisition module can achieve the effect of simultaneous sampling of four channels of data, so the data compression module also needs to ensure the simultaneous compression of four channels of data. After each channel of data is buffered in FIFO, it is input into the wavelet transform data compression module for wavelet transform. The compression results are all 20MHz frequency signals. After the beat is a 40MHz frequency signal output, two flag bits are added to determine the channel to which it belongs. The synchronous output results of the four channels are spliced ​​into a 40MHz frequency 64-bit signal and stored in the DDR cache. When the reserved bits of d1, d2, and d3 are all set to 6 bits, the 56-bit single-channel splicing result needs to be divided into 4 14-bit outputs, and then the flag bit is added; when the reserved bits of d1, d2, and d3 are all zero, the flag bit is directly added.

[0065] Step 5: Calculate the compression ratio of the compression module by the ratio of the data size before and after signal compression. According to the compression ratio calculation formula, when the reserved bits of d1, d2 and d3 are all set to 6 bits, the ratio is 2, and when the reserved bits of d1, d2 and d3 are all zero, the compression ratio is 8.

[0066] Step 6: The compressed and spliced ​​results are uploaded to the host computer 107 through the data transmission module 106 for channel demultiplexing to obtain the output compression results of the independent 60-channel detectors. The reconstructed signal can be obtained through data de-splicing and wavelet reconstruction.

Claims

1. An absorption spectrum tomography system based on channel multiplexing and real-time data compression, comprising a laser generating module, a tomography sensor, a data acquisition module, a channel multiplexing module, a data compression processing module, a data transmission module and a host computer, etc.; the laser generating module controls the tunable diode laser in the tomography sensor to output scanning laser in a time-division multiplexing manner, and couples into a beam of light by an optical fiber coupler, which is collimated by a collimator on a fan-beam sensor after optical fiber transmission, and is widened into a fan-shaped laser beam by a Powell prism. In order to avoid mutual interference of light intensity signals at different angles, each laser in the tomography sensor is output in a serial scanning manner. The channel multiplexing module groups the detectors according to the laser opening timing in the tomography sensor. Each group contains two detectors. The useful signal timing of the two detectors is staggered and connected to the same acquisition channel of the data acquisition module. The parallel data acquisition module samples the uncompressed absorption spectrum data. The data compression processing module completes the on-chip wavelet basis decomposition, bit quantization and data splicing to obtain the wavelet coefficients suitable for storage, so as to realize real-time data compression. The wavelet coefficients stored in the on-chip cache chip after compression processing are uploaded to the host computer through the data transmission module, and the reconstructed absorption spectrum signal is obtained in the host computer after data de-splicing and data reconstruction.

2. The absorption spectrum tomography system based on channel multiplexing and real-time data compression according to claim 1, characterized in that: The system design is based on a channel multiplexing module of the time-sharing scanning mode of the tomography imaging sensor. Combined with the laser opening timing in the sensor, the detector grouping and connection method are designed to multiplex two detector outputs into one acquisition input to maximize the duty cycle of the acquisition system. The structure of the tomography sensor is designed as a regular N-gon, that is, it contains 2*N tunable diode lasers. The laser generation module controls the tunable diode lasers in the tomography sensor to output scanning lasers in a time-division multiplexing manner. The laser output by each laser is coupled into a beam of light using an optical fiber coupler, which is collimated by a collimator on the fan beam sensor after optical fiber transmission, and widened into a fan-shaped laser beam by a Powell prism, with a beam expansion angle of β°; the output laser passes through the gas to be measured in the area of ​​interest of the tomography sensor, and the transmitted laser is received by y photodetectors at the same time. The beam expansion angle θ° and the number of detectors y that simultaneously receive effective signals can be determined according to the size of the area of ​​interest; if a detectors are installed on each side of the sensor, the total number of detectors Y contained in the tomography sensor of N angles is N*a, y needs to satisfy the conditions of a≤y≤2a, and the left and right numbers are symmetrical, which can ensure that when there is a large area of ​​interest, the laser signals at each angle do not interfere with each other, then each detector will receive laser signals at two angles at this time; if the scanning time of a single-angle laser signal is T r , the time for each detector to receive the signal is 2*T r , the time required to scan all N angles is N*T r , then the duty cycle of the detector output signal is When Y is an even number, the detector is divided into Each group has two detectors and they do not receive valid signals at the same time. The two detectors in each group are connected to the same acquisition channel. That is, when laser 1 is working, the corresponding y detectors are in working state, and the other detectors are in non-working state. According to the number of sensor angles, two groups of detectors that do not interfere with each other when lasers at different angles are turned on can be divided. They are connected to the same group of acquisition channels in the scanning order to realize the multiplexing of the acquisition channels and improve the duty cycle of the system to When Y is an odd number, the detector is divided into Group, front Each group has two detectors, and the two detectors in each group are connected to the same acquisition channel. The last group has only one detector, which is connected to one acquisition channel alone. The duty cycle of the system is still increased to 3. The absorption spectrum tomography system based on channel multiplexing and real-time data compression according to claim 1, characterized in that: The system designs an on-chip data compression processing module, selects the field programmable gate circuit device FPGA as a processor, performs wavelet decomposition on the transmitted laser intensity signal to solve all detail coefficients and approximate coefficients in real time according to the selected wavelet basis function, decomposition layer number and threshold range and other parameters as the initialization parameters of the data compression processing module, performs threshold processing and quantization processing on the coefficients in the FPGA, and then splices them into a signal form suitable for storage, and finally, the compressed signal is de-spliced ​​and wavelet reconstructed by the host computer to obtain a reconstructed signal. The detailed process is carried out according to the following steps: The first step is to select wavelet basis function, decomposition layer number and threshold range as initialization parameters of data compression processing module; The original signal is decomposed into a single layer using the lifting wavelet transform method to obtain approximate coefficients and detail coefficients. The wavelet transform method is: in, and ψ j,k (t) are called scaling function and wavelet function respectively, A j,k and D j,k are the approximate coefficient and the detail coefficient. The multi-resolution analysis algorithm is introduced, the decomposition layer number j is preset, the detail coefficient of each layer is thresholded, and a new detail coefficient is obtained. The approximate coefficient of the last layer is retained, and each coefficient is reconstructed by wavelet. The decomposition layer number is changed, and the signal-to-noise ratio of the reconstructed signal under different decomposition coefficients is calculated. The layer number with the smallest signal-to-noise ratio is selected to obtain the denoised absorption spectrum signal, and the wavelet basis function is changed to repeat the above operation, and the reconstruction results are compared to select the appropriate solution. The three parameters of wavelet basis function, decomposition layer number and threshold range are retained as the initialization parameters of the data compression processing module. The second step is to use the FPGA processor in the data compression processing module to perform on-chip wavelet decomposition on the sampled signal, calculate all detail coefficients and approximate coefficients in real time, and perform threshold processing and quantization processing on the required coefficients in the FPGA; The parallel data acquisition module quantizes the channel multiplexed transmitted laser intensity signal, and the FPGA processor controls it to send the digital signal of the laser intensity to the data compression processing module in real time to realize on-chip wavelet transform; the FPGA performs streaming parallel processing on the input data, that is, it does not need to cache the data or intermediate results in the external dynamic random access memory DRAM for buffering, but only uses on-chip memory resources. This streaming design ensures high throughput and low latency of the processing process; in the data compression processing module, the signal is calculated in binary data mode, and the data selection control signal and multiple registers are set. The selection signal is used to control the sequence selection of the register to select the subsequent calculation order and the output data order to complete the prediction and update steps of the lifting wavelet transform; all detail coefficients and approximate coefficients of the multi-layer wavelet transform are obtained in sequence to realize the on-chip implementation of the lifting wavelet transform; the hardware design of this method is relatively simple, resource-saving, and flexible to use; only the multi-layer detail coefficients and the last layer of approximate coefficients of the demodulated original absorption spectrum signal are retained, and the number of bits of the retained coefficients is quantized, and the number of bits of the coefficients is reduced without affecting the range of the coefficient value, so as to realize the data compression function; The third step is to perform a single-channel data splicing operation on the compressed coefficients; Since the multi-layer wavelet transform hardware implementation process needs to continuously perform frequency division operations, each time a layer of wavelet transform is performed, the main clock frequency of the wavelet transform of this layer is half of the main clock frequency of the wavelet transform of the previous layer. Therefore, splicing processing is required before the quantized coefficients are stored in the subsequent storage; first, after sampling by the analog-to-digital converter, an absorption spectrum signal with a point number of n and a clock frequency of v is obtained. The frequency of the detail coefficients and approximate coefficients obtained after the first layer of wavelet transform is v1, and the number of points is The frequency of the detail coefficients and approximate coefficients obtained after the second layer of wavelet transform is v2, and the number of points is The frequency of the detail coefficients and approximate coefficients obtained after the j-th layer wavelet transform is v j , the points are Then the frequencies after decomposition of each layer satisfy the following relationship: v=2*v1=2 2 *v2=···=2 n *v j (2) Then, the number of bits of the detail coefficients decomposed at the first layer is quantized to x1, the number of bits of the detail coefficients decomposed at the second layer is quantized to x2, and the number of bits of the detail coefficients decomposed at the jth layer is quantized to x j , the number of bits of the approximate coefficients obtained by the last layer of decomposition remains unchanged and is consistent with the number of bits of the original data, which is x; finally, the clock domain is unified by beating, and the coefficients obtained after compression are aligned, with the period of the coefficient with the smallest frequency as the alignment unit. Then the spliced ​​binary result l1 of the single-channel acquisition of n absorption spectrum data after compression is: Step 4: realize data splicing of acquisition modules containing multiple acquisition channels; One acquisition module is designed with four acquisition channels, so the data compression module also needs to ensure the simultaneous compression of four channels of data. After each channel of data enters the FIFO buffer, it is input into the wavelet transform data compression module for j-layer wavelet decomposition. The splicing results of the four channels are obtained according to the number of quantization bits. The quantization and splicing methods of each channel are consistent. It is necessary to add a two-bit flag prefix to the splicing result of each channel. Channel 1 is 00, channel 2 is 01, channel 3 is 10, and channel 4 is 11. The binary results after adding the flag are l1′, l2′, l3′ and l4′. The output results are independent of each other and the clock domain is consistent. The splicing result of the four channels is that the first bit of each channel after adding the flag is connected: l＝[l1′，l2′，l3′，l4′] (4) Step 5: Calculate the compression ratio CR of the compression module by the ratio of the data size before and after signal compression: According to the above splicing results, the actual compression ratio of the data compression module can be calculated as: The actual compression ratio depends on the set quantization bit number and the number of decomposition layers. The smaller the quantization bit number and the higher the number of decomposition layers, the greater the compression ratio. j =0, the compression ratio reaches the maximum value; Step 6: Convert the spliced ​​compression result into a data format suitable for writing into the storage chip. After caching in the storage chip, the data transmission module uploads all the data to the host computer to de-splice the compressed signal and reconstruct it by wavelet to obtain the reconstructed signal.

Images