Open Access
Add to BibSonomyAbstract
Emission spectral tomography (EST) has been adopted to test the three-dimensional distribution parameters of fluid fields, such as burning gas, flame and plasma etc. In most cases, emission spectral data received by the video cameras are enormous so that the emission spectral tomography calculation is often time-consuming. Hence, accelerating calculation becomes the chief factor that one must consider for the practical application of EST. To solve the problem, a hardware implementation method was proposed in this paper, which adopted a digital signal processor (DSP) DM642 in an emission spectral tomography test system. The EST algorithm was fulfilled in the DSP, then calculation results were transmitted to the main computer via the user datagram protocol. Compared with purely VC++ software implementations, this new approach can decrease the calculation time significantly.
©2009 Optical Society of America
1. Introduction
Generally, Optical computed tomography (OpCT) including interferometer [1], Moire grating [2], and light beam deflection [3] was adopted to measure the three-dimensional distribution parameters of fluid fields. Compared with these OpCT techniques, an emission spectral tomography (EST) system does not need laser resources; hence its structure is simple and more practical [4]. In the early stage of EST applications, the emission intensity data were obtained and transmitted to the spectral analysis systems through optical-fiber scanning modules. The scanning process is often time-consuming, thus these systems can only be used to test stable fluids like the arc argon plasma [5, 6]. To test transient fluids with EST, a multichannel video CCD test system was proposed [7], through which multi-orientation spectral intensity images can be obtained simultaneously to ensure the real-time data receival. However, in the case of online test, the data obtained are often so large that the tomography reconstruction calculation becomes the key factor to meet the real-time demand. A few timesaving and efficient OpCT algorithms were proposed; however, those studies only focused on optimizing the flow of program and seldom considered the implementation methods of these algorithms [8, 9]. Digital signal processor (DSP) has underwent fast development during recent years, and some DSPs have specific designed structures that aimed at video signal processing [10]. If EST reconstruction calculation were implemented with these DSPs, it would be timesaving. An online EST system with a 4-channel video signal processor DM642 was proposed in this paper and the DSP inner software framework was also introduced in detail. Comparison studies showed that this new approach had better real-time performance than conventional purely VC++ software implementations.
2. Implementation procedure
2.1 Hardware structure
In our prior study for EST applications, an experimental system was employed [7]. Here, based on this system, an online EST system with DM642 was proposed, which is illustrated in Fig. 1. The tested object, a flame of a candle, was located in the center of a circular platform and eradiated toward all directions in the space. Four industrial video cameras adopted to capture continuous spectral intensity images were placed around the center of the platform and along the axes dividing the total angles into four even parts. The two-dimensional spectral intensity images I of the central frequency v of narrowband optical filters are related to not only the radial position x’ but also the azimuth angle ϕ. As shown in Fig. 2, the relationship between the emission coefficient ε(v) with the spectral intensity I(v) is
Therefore, the spectral intensity is the projection (line integral) of the emission coefficient of the flame along the coordinate axis y’. Namely, I(v) is the Radon transform of ε(v). Here, I(v) was sensed by the charge-coupled device (CCD) of the video cameras via the collimation tubes. The four-channel real-time data were transferred to a DSP-based circuit fulfilling the EST reconstruction calculation and the final reconstruction results were transmitted to the central computer.
The DSP-based EST reconstruction circuit as shown in Fig. 3 consists of a 32-bit digital signal processor TMS320DM642 with working frequency 720MHz, four video decoders TVP5150PBS decoding the spectral intensity data and giving decoded data to DM642, a 32M-byte synchronous dynamic random access memory (SDRAM) storing the original digital video data before reconstruction and calculated data after construction, a 4M-byte flash memory fixing DSP programs, a 32-bit PCI 2.2 bus, and a 10/100M ethernet interface RJ45. The circuit was connected to a main computer via interface RJ45 and the data was transmitted via the user datagram protocol (UDP). At the beginning of the test, the main computer program sent the reconstruction parameters to and then initiated the EST reconstruction circuit. The DM642 finished the reconstruction calculation and then final results were sent to main program via UDP to be stored or shown in figures.
2.2 Software design
The EST reconstruction problem of the emission coefficient distribution ε(x,y,z,v) can come down to the inverse Radon transform that can be described based on the series expansion principle as
Where I is the measurement matrix consisting of I(v) of different directions; W is the projection matrix; E which consists of the constructed emission coefficient ε(v) is the image matrix; σ is the measurement errors matrix including random errors in the testing process and the inherent errors of the testing method. Resolving Eq. (2) equals to estimating the image matrix Ê depending on measurement matrix I and certain optimization criterion.
The EST reconstruction algorithm employed in our system is SPDA-MCIR [7] that is based on the multi-criterion optimization principle and has a self-adaptive filtering prior to the iterations. The SPDA-MCIR iterative formula is
where M, N is the reconstructed pixel along x, y axis respectively; d is the product of the total testing directions and the number of rays per view (RPV); a matrix B’ is introduced in iteration to satisfy the minimum norm and the maximum average cretirion [6]; γ is the relaxation parameter and λ1(k),λ2(k), λ3(k) are the weigh factors corresponding to the proportions of different criterions that are self-adaptively adjusted in iterations in the SPDA-MCIR algorithm.
This algorithm must be transformed into DSP language and be implemented by the DSP. A code composer studio2.2 tool offered by the DSP producer enables us to transform C++ resource programs to DSP language. Besides, the key of inner DSP program design is synchronization and optimization. The EST reconstruction requires the multi-channel original spectral intensity data to be captured synchronously. Synchronous extraction of the multichannel data needs to be considered in the DSP program. The DSP/BIOS real-time core was employed and the synchronous module of the reference framework level 5 (RF5) was adopted for timing of the software parts, whose flowchart is shown in Fig. 4. The purpose of optimization is to accelerate calculation. In the process of compiling, the linear compiling and the pipeline optimization manner were applied.
The main computer program was developed with VC++, which included reconstruction parameters setting, UDP communication, reconstructed data receival, store, and display, etc.
3. Experimental results
To compare the calculation speed of pure software and the DSP-based approach, four local spectral intensity images (150 pixels multiplied by 51 pixels) of the tested flame in four evenly spaced azimuthal directions at a certain instant, as shown in Fig. 5, were extracted to reconstruct the distribution of emission coefficients εv of the flame. Comparison between reconstruction time of software and that of DSP-based hardware implementations with the SPDA-MCIR algorithm is shown in Table 1, where the reconstructed region of whole flame was divided into nine cross sections and the reconstruction resolution of each cross section was variable. The reconstruction results of the DSP-based approach were transmitted to the main computer program to be displayed via the UDP protocol. The reconstructed three-dimensional distribution of emission coefficients of the flame is illustrated in Fig. 6.
Table 1. Comparison of reconstruction time of VC++ software and DSP-based approach (SPDA-MCIR algorithm, iteration times=100, relaxation parameter=0.009)
4. Conclusion
Comparison experiments showed the new approach based on DSP642 could decrease the calculation time significantly than those based on purely VC++ software implementations. Furthermore, the improvement becomes more obvious when the reconstruction size increases. Hence, this emission spectral tomography (EST) system with DSP is promising in the online test. Futhermore, with the development of DSPs technology similar approaches based on DSPs will be widely applied in the optical test where time-consuming calculation is needed.
Acknowledgments
The project is supported by Chinese Natural Science Foundation (grant 60577016), Jiangxi Natural Science Foundation (grant 0512034), Aeronautical Science Foundation (grant 2006ZD56004), Key Laboratory Foundation of Jiangxi Education Bureau (grants 2005-314 and 2006-164) and Doctoral Foundation of Nanchang Hangkong University.
References and links
1. S. Bahl and J. A. Liburdy, “Three-dimensional image reconstruction using interferometric data from a limited field of view with noise,” Appl. Opt. 30, 4218–4226 (1991). [CrossRef] [PubMed]
2. Y. S. Cheng, “Two-dimensional grating interferometric imaging by computed tomography,“ Opt. Lett. 12230–233 (1987). [CrossRef] [PubMed]
3. Yiqing Gao, XingDao He, and Yongqing Gong, “Radon transform iteration based on beam-deflection optical tomography,” Proc. SPIE 4221, 274–278(2000). [CrossRef]
4. L. I. Poplevina, I. M. Tokmulin, and G. N. Vishnyakov, “Emission spectral tomography of multijet plasm flow,” in Inverse Optics III, M. A. Fiddy, ed., Proc. SPIE 2241, 90–98 (1994). [CrossRef]
5. M. Hino, T. Aono, M. Nakajima, and S. Yuta, “Light emission computed tomography system for plasma diagnostics,” Appl. Opt. 26, 4742–4746 (1987). [CrossRef] [PubMed]
6. X. Wan, S. Yu, G. Cai, Y. Gao, and J. Yi, “Three-dimensional plasma field reconstruction with multiobjective optimization emission spectral tomography,” J. Opt. Soc. Am. A 21, 1161–1171 (2004). [CrossRef]
7. X. Wan and A. Yin, “Denoising multicriterion iterative reconstruction in emission spectral tomography,“ Appl. Opt. 46, 1223–1232 (2007). [CrossRef] [PubMed]
8. Jue Wang, Yan-Ping Lu, and Yu-Fang Cai, “The application of volumetric region growing in segmentatio for volume data from industrial computed tomography,” Proc. SPIE 6041, 112–117 (2005).
9. Jonathan S. Maltz, “Multiresolution constrained least-squares algorithm for direct estimation of time activity curves from dynamic ECT projection data,” Proc. SPIE 3979, 586–598 (2000). [CrossRef]
10. V. Brost, S. Bouchoux, F. Yang, M. Paindavoine, and J. C. Grapin, “Real-time implementation of face tracking on DSP TMS320C6x embedded system,” Proc. SPIE 4948, 701–706 (2003). [CrossRef]
