Abstract
We propose a compression-coding-based surface measurement method that combines single-pixel imaging and heterodyne interference using an optical frequency comb. The real and imaginary parts of the heterodyne interference signals are used to obtain the depth information rapidly. By optimizing the ordering of the Hadamard measurement basis, we reconstruct a three-step sample with heights of approximately 10, 20, and 30 µm without an iterative operation in 6 ms, with a precision of 5 nm. Compared with the uncompressed measurement, the sampling times reduced to 20%, and the measurement time reduced by five times without measurement accuracy loss. The proposed method is effective for rapid measurements, particularly for objects with a simple surface topography.
© 2021 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
1. Introduction
An optical frequency comb (OFC) emits a series of ultrashort optical pulses with equal time intervals. In the frequency domain, the Fourier spectrum is composed of many optical longitudinal modes that are equidistantly spaced at the repetition frequency [1–4]. These optical longitudinal modes can be precisely phase locked to a frequency standard. Hence, the OFC can be regarded as an optical frequency ruler and applied extensively in diverse fields, including optical clocks [5,6], absolute distance measurement [7–9], refractive index measurement [10,11], angle measurement [12–14] and surface measurement [15–23]. In the field of high-precision surface measurement, the OFC can be used to perform profilometry with a higher accuracy and a larger range compared with the traditional light source. Hence, a series of measurement methods has been proposed, such as repetition frequency scanning [15,17], chirped pulses [16], spectral interferometry [24], mechanical scanning [25], and single-pixel-imaging (SPI) [18–20]. Among these methods, the SPI-combined OFC surface measurement method has garnered significant attention owing to its characteristics of lower cost, faster measurement speed, and no additional mechanical vibration, afforded by the use of a high-speed modulated digital micromirror device (DMD).
In the SPI surface measurement system, a series of coding masks for sampling and the corresponding reconstruction algorithm are essential. The sampling and reconstruction methods significantly affect the reconstruction speed and quality [18–20,26,27]. The first approach is based on compressive sensing (CS). By constructing a sparse measurement signal and solving the underdetermined equation sets, one can reconstruct the profile of an object [18,20,26]. This CS-based method can achieve compression measurements with a sampling rate lower than the Nyquist limit [20,28,29], i.e., the object’s profile is reconstructed completely at a sampling times less than the number of measuring points. However, the CS reconstruction process is time consuming because an iterative algorithm is required to solve the convex optimization problem. In addition, sparse measurement signals cannot be structured easily for complex surface profiles. The other approach is based on cyclic Hadamard transform imaging. By inverse matrix transformation, the object’s profile can be reconstructed rapidly without iteration [19,27]. This method affords high reconstruction efficiency, and the modulation matrix is binary, which is easy to use on high-speed spatial light modulators. However, the sampling times must be equal to the measuring point to reconstruct the object’s profile, hence, compression sampling cannot be realized.
To achieve effective compression sampling and rapid reconstruction, the SPI surface sampling and reconstruction method must be optimized. For SPI methods, alternative compression approaches have been proposed for real-time imaging with a tradeoff between image quality and real-time robustness, such as evolutionary compressive sensing [30–32] and optimized ordering of the Hadamard basis [33,34]. Because different bases contribute differently to image reconstruction, in these methods, more significant basis masks are projected and measured to reduce the sampling times. Moreover, for the abovementioned methods, the detected value of the single-pixel detector must satisfy the linear relationship with the measuring values at all measuring points. In the OFC-based surface measurement, the depth information is obtained by solving the phase of the longitudinal mode, and the phase of the signal detected by a single-pixel detector is typically nonlinear to the phase of each measuring point. Therefore, the nonlinear correlation between the phases of the detection signal and the measuring points must be solved to apply the optimized ordering of the Hadamard basis in the OFC surface measurement. Additionally, an appropriate ordering of the Hadamard basis must be constructed for surface measurements.
In this article, we propose a compression coding-based surface measurement method using SPI and OFC heterodyne interferometry. Using the fast Fourier transform (FFT) for the heterodyne interference signal (HIS) recorded by the single-pixel detector, we can obtain the real and imaginary parts at the beat frequency that exhibit a linear relationship with the real and imaginary parts of the HIS with all measuring points, respectively. Additionally, owing to the additional effect of the wavefront error on the measuring surface, it is difficult to effectively perform compression with a fixed ordering of Hadamard basis. To improve the sampling compression ratio, an adaptive optimized ordering of the Hadamard basis is applied to reconstruct the surface profile without a complex wavefront correction. Consequently, the measuring time of a three-step profile improves to several milliseconds with nanoscale precision.
2. System configuration
Figure 1 shows the SPI-based OFC heterodyne interference system, which illustrates the process by which the SPI scheme of the optimized ordering of Hadamard basis is incorporated into the OFC heterodyne interference scheme. The light source is a mode-locked Er-doped fiber laser with a central wavelength of 1.56 µm and a spectral bandwidth of 53 nm. The repetition frequency fr is ∼77.8 MHz, which is phase locked to a rubidium frequency standard (SIM940, Stanford Research Systems, USA, accuracy, 5 × 10−11; instability, 2 × 10−11 at 1 s) by controlling the laser cavity length using a piezoelectric transducer and a temperature-controlled box. The pulse train from the light source is segregated into two paths by a 50:50 beam splitter and served as reference and measurement pulses, respectively. A beam expander, which is composed of a micro-objective, pinhole, and collimating lens, is inserted in the measurement path to expand the measurement beams before they reached the measurement surface. The beam expander in the reference path has the same amplification parameters with the measurement path, and an acousto–optic modulator (AOM) is inserted into the reference path to shift the optical frequency prior to beam expansion. To construct a heterodyne interference signal with frequency fb, the shift frequency (fs) by the AOM is set to fs = fr + fb in our system, furthermore, the fs is also referenced to the rubidium frequency standard to ensure the stability of the heterodyne frequency. The measurement beam carrying information pertaining to the target surface is superimposed with the reference beam and results in heterodyne interference at BS2. The heterodyne interference beam is segregated into two beams by BS3. One beam is detected by a photodetector (PD2) after lens focusing, and the other beam, modulated by a DMD (DLP6500, Texas Instruments), is focused to another photodetector (PD1). The DMD’s reflecting surface contained more than 2 million micromirrors with a side length of 7.56 µm, and each micromirror has two operating states with the highest modulation frequency over 9.5 kHz corresponding to different reflection angles. The reflected light corresponding to one of the operating states is reflected to PD1, whereas the other reflected light will not be received in our system. The single-pixel photodetectors (PDA20CS2) with a bandwidth of 11 MHz and an active area of 3.14 mm2 (Ø2.0 mm) ensures that the high-frequency heterodyne signals are adequately sampled and fully received. Both heterodyne interference signals recorded by PD1 and PD2 are sampled synchronously using a data acquisition card. We measure a compensate flat surface in advance to eliminate wavefront error caused by different beam expansions and optimize the ordering of Hadamard basis. To improve the longitudinal measurement range of the surface measurement system, two band-pass filters with different central wavelengths are alternatively inserted into the system to construct the synthetic wavelength [8,17,35]. The inset in Fig. 1 illustrates the filtered spectra.
3. Principles of compression-coding-based surface measurement
Figure 2 shows the compression-coding-based surface measurement processing using the SPI of the optimized ordering of Hadamard basis and the OFC heterodyne interferometry. Owing to the equal intervals and the linearly dependent stable phases of the OFC longitudinal modes [36], the HIS containing the information of the sample surface can be expressed as shown in Eq. (1) before being modulated by the DMD.
where Ii is the heterodyne interference signal corresponding to the ith measuring point; Ai, fb, and φc,i are the amplitude, beat frequency, and phase of the heterodyne interference signal, respectively. The phase of the heterodyne interference signal is equivalent to that of the effective central wavelength of the optical spectrum based on the previous study [37].The coding masks (CMs) loaded and displayed on the DMD are based on an N × N Hadamard matrix H. Every CM is adjusted from a one-dimensional to two-dimensional n × n distribution, where N = n × n. Moreover, the CM unit corresponding to each measuring point is composed of many micromirrors with the same operating state within the unit areas. The binary elements of the CM correspond to states reflecting or not reflecting light to PD1, respectively. After being modulated successively by the DMD, the HIS is expressed as
where ${Y^j}$ is the HIS detected by PD1 in the jth measurement; “1” and “0” of w correspond to passing and blocking light, respectively. To reduce the effect of air disturbance, part of the HIS is detected at PD2 as the reference HIS. Based on the Fourier transform of the HIS and the characteristics of the H matrix, the real parts of the HIS based on the reference signals for all CMs are expressed asBecause the Hadamard matrix is a symmetric orthogonal matrix (H = HT + H−1), the surface profile can be reconstructed perfectly by the inverse matrix transform when all bases are used for the measurement [19,38]. To reduce the measurement time and realize compression measurements, the real and imaginary parts of the measuring points can still be restored using partial bases. However, the reconstruction quality is related to the importance of partial measurement bases [30,32]. Therefore, the ordering of the measurement bases significantly affects the reconstruction efficiency when using partial bases for measurement [33]. In our system, the measurement results contain both the sample surface profile information and the wavefront error between the reference and measurement arms. The wavefront error can be determined by measuring the standard compensating plate, and the measurement bases are sorted in the order of priority for the error reconstruction. For some simple surfaces, the necessary bases are typically contained within the bases for wavefront error reconstruction. Therefore, rapid measurements can be realized effectively by reconstructing the wavefront error rapidly. To evaluate the importance of the measurement bases for reconstructing the wavefront error, all the measurement bases are used to measure the compensate plane and obtain the real and imaginary parts of the modulated HIS. The ordering factor can be calculated as follows:
Subsequently, the distance from the ith measuring point can be expressed as
To expand the measuring longitudinal range, two filters with different central wavelengths are used to form the synthetic wavelength alternately. When the precision of the synthetic wavelength measurement is less than a quarter of a single wavelength, the distance can be precisely determined using a single wavelength [17].
4. Experiments and results
To evaluate the performance of the optimized ordering of the Hadamard basis, we derived an object’s profile using this compression-coding-based surface measurement, with N = 256 and n = 16, and the size of the unit pixel of the CM is 378 µm × 378 µm, which is equal to the spatial resolution due to a parallel light path in the equal arm interferometer. Figure 3(a) shows a photograph of a three-step sample comprising three gauge blocks of 1.23, 1.24 and 1.26 mm thickness respectively with different heights of 10, 20, and 30 µm. After measuring the compensated surface and optimizing the ordering of the measurement basis using the procedure described above, we measured the three-step sample in area II (6.048 mm × 6.048 mm) with a blue square box, as shown in Fig. 1(a). We used the root mean square (RMS) of the error between the compression measurement and the all-bases measurement to evaluate the performance of the compression measurement; it is calculated as
Next, we conducted a comparative experiment using a commercial white light interferometer (WLI, Bruker, Contour GT-IM) with a step height accuracy of 0.75% to validate the height accuracy. The measurement results are shown in Fig. 4(a) for an area measuring 1.3 mm × 0.9 mm, which was limited by the field of view of WLI, and the step heights corresponding to the two reference lines were 20.049 and 29.962 µm, separately. Figure 4(b) shows the results reconstructed via the OFC-based compression measurement at a 20% sampling rate. Considering the reference lines at the same positions, the step heights were 29.990 µm (cross-section A) and 20.132 µm (cross-section B), which are consistent with the results of WLI. It was verified that this method can achieve nanoscale precision in measuring the step height.
The experiments conducted show that a basal number of measurement bases is required to reconstruct the measuring surfaces effectively. Different surface morphologies require different number of the measurement bases. To validate the effect of surface complexity on the reconstruction quality, we conducted a comparative experiment for different surfaces which are parts of the MEMS device having a step-shaped platform and a teardrop-shaped groove, separately. The nominal height of both parts was approximately 10 µm. The reconstructed results at a 20% sampling rate for the step-shaped platform and teardrop-shaped groove are shown in Figs. 5(a) and 5(b), respectively, which are consistent with the actual morphology of the objects. Figure 5(c) shows the RMS error between the compression and all-bases measurements for the measuring objects, where the blue and red lines correspond to the results of the step-shaped and teardrop-shaped MEMS devices, respectively. The RMS error curve of the teardrop-shaped surface decreased more slowly than that of the step-shaped surface. This shows that the reconstruction of a more complex surface was less efficient than that of a simple surface. The experimental results prove that the proposed compression coding surface measurement is effective for different surfaces. Nevertheless, a higher sampling rate is preferred for complex surface measurements. Hence, our compression measurement method is more suitable for sparse surfaces, and the compression efficiency is limited by the complexity of the surfaces.
5. Discussion and conclusion
The proposed scheme of optimized ordering of the Hadamard matrix was verified to achieve high-precision single-pixel compression-coding-based surface measurements. Compared with CS technology, the reconstruction processing is more efficient without requiring iterations, which reduces the reconstruction time at a permissible loss of the reconstruction quality. In optimizing the ordering of the Hadamard basis, some efficient ordering methods include the Walsh–Hadamard ordering [33], which is suitable for recovering images of low spatial frequency because the bases for low spatial frequency are always used preferentially. However, owing to the wavefront error in the surface measurement, the measuring surface is always a complex surface, even if the sample is simple. Our adaptive optimization method orders the measurement bases based on the surface error and can effectively compress the sampling rate without strict wavefront alignment; Furthermore, the fceo can be free-running. Because the experimental upset is based on an equal arm interferometer and the overlapping pulses from the measurement and reference arms have the same pulse number. Therefore, the phase difference between the measurement and reference pulses doesn’t change with the fceo. However, the measurement speed of this method is limited by the modulation rate of DMD and the measurement speed will increase as the development of the DMD modulation rate. In addition, the spatial resolution is limited by the light intensity. By using a stronger intense light source, less micro-mirrors are needed to be used as an element and the spatial resolution is expected to enable a high lateral resolution up to the limitation of a micro-mirror size.
In summary, we demonstrate a compression-coding-based surface measurement method using SPI and OFC heterodyne interferometry. By optimizing the ordering of the Hadamard basis, the phase of the heterodyne interference signals can be reconstructed rapidly through the reconstruction of real and imaginary parts. A group of synthetic wavelengths of the OFC is used to extend the measurement depth, and a three-step surface with heights of 10, 20, and 30 µm is measured in 6 ms at a sampling rate of 20% with a precision of 5 nm. The experimental results showed that the surface profile can be reconstructed at a sampling rate lower than the Nyquist limit with an acceptable loss of reconstruction quality, and that the reconstruction is fast because no iteration is required. In addition, the measurement bases are contained in a binary Hadamard matrix that is generated rapidly by the Hadamard transform, thereby avoiding the storage of many matrices and easing their use in a DMD with a fast binary modulation rate. Our proposed method based on the OFC and SPI is feasible for rapid surface measurements, particularly for objects with a simple surface topography.
Appendix
A. Construction of real parts of modulated signals
Figure 2 shows the compression-coding-based surface measurement procedures using SPI with the optimized ordering of the Hadamard bases and the OFC heterodyne interference. When a series of CMs is displayed on the DMD, the initial phase of the modulation HIS is affected by the unstable modulation intervals. To reduce the effects of air disturbance and determine the effective initial phase, a portion of the HIS is assigned to the PD2 and sampled synchronously with PD1 as the reference HIS. Both the amplitude and phase of the modulated and reference HIS at frequency fb can be obtained via FFT, and the relative complex value of the effective signal can be expressed as
Funding
Tsinghua University; Beijing Science and Technology Planning Project (Z191100007419011); National Natural Science Foundation of China (51835007).
Disclosures
The authors declare no conflicts of interest.
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
References
1. J. Ye and S. T. Cundiff, Femtosecond Optical Frequency Comb: Principle, Operation and Applications (Springer US, 2005).
2. T. Udem, R. Holzwarth, and T. W. Hänsch, “Optical frequency metrology,” Nature 416(6877), 233–237 (2002). [CrossRef]
3. T. Udem, J. Reichert, R. Holzwarth, and T. W. Hänsch, “Accurate measurement of large optical frequency differences with a mode-locked laser,” Opt. Lett. 24(13), 881–883 (1999). [CrossRef]
4. D. J. Jones, S. A. Diddams, J. K. Ranka, A. Stentz, R. S. Windeler, J. L. Hall, and S. T. Cundiff, “Carrier-envelope phase control of femtosecond mode-locked lasers and direct optical frequency synthesis,” Science 288(5466), 635–639 (2000). [CrossRef]
5. M. Takamoto, F. L. Hong, R. Higashi, and H. Katori, “An optical lattice clock,” Nature 435(7040), 321–324 (2005). [CrossRef]
6. B. J. Bloom, T. L. Nicholson, J. R. Williams, S. L. Campbell, M. Bishof, X. Zhang, W. Zhang, S. L. Bromley, and J. Ye, “An optical lattice clock with accuracy and stability at the 10(−18) level,” Nature 506(7486), 71–75 (2014). [CrossRef]
7. H. Y. Zhang, H. Y. Wei, X. J. Wu, H. L. Yang, and Y. Li, “Absolute distance measurement by dual-comb nonlinear asynchronous optical sampling,” Opt. Express 22(6), 6597–6604 (2014). [CrossRef]
8. Z. Zhu, G. Xu, K. Ni, Q. Zhou, and G. Wu, “Synthetic-wavelength-based dual-comb interferometry for fast and precise absolute distance measurement,” Opt. Express 26(5), 5747–5757 (2018). [CrossRef]
9. K. N. Joo and S. W. Kim, “Absolute distance measurement by dispersive interferometry using a femtosecond pulse laser,” Opt. Express 14(13), 5954–5960 (2006). [CrossRef]
10. G. Wu, M. Takahashi, K. Arai, H. Inaba, and K. Minoshima, “Extremely high-accuracy correction of air refractive index using two-colour optical frequency combs,” Sci. Rep. 3(1), 1894 (2013). [CrossRef]
11. P. Balling, P. Mašika, P. Křen, and M. Doležal, “Length and refractive index measurement by Fourier transform interferometry and frequency comb spectroscopy,” Meas. Sci. Technol. 23(9), 094001 (2012). [CrossRef]
12. S. Zhou, V. Le, S. Xiong, Y. Yang, K. Ni, Q. Zhou, and G. Wu, “Dual-comb spectroscopy resolved three-degree-of-freedom sensing,” Photonics Res. 9(2), 243–251 (2021). [CrossRef]
13. X. Liang, J. Lin, L. Yang, T. Wu, Y. Liu, and J. Zhu, “Simultaneous Measurement of Absolute Distance and Angle Based on Dispersive Interferometry,” IEEE Photonics Technol. Lett. 32(8), 449–452 (2020). [CrossRef]
14. Y.-L. Chen, Y. Shimizu, J. Tamada, Y. Kudo, S. Madokoro, K. Nakamura, and W. Gao, “Optical frequency domain angle measurement in a femtosecond laser autocollimator,” Opt. Express 25(14), 16725–16738 (2017). [CrossRef]
15. W.-D. Joo, J. Park, S. Kim, S. Kim, Y. Kim, S.-W. Kim, and Y.-J. Kim, “Phase shifting interferometry for large-sized surface measurements by sweeping the repetition rate of femtosecond light pulses,” Int. J. Precis. Eng. Manuf. 14(2), 241–246 (2013). [CrossRef]
16. T. Kato, M. Uchida, Y. Tanaka, and K. Minoshima, “High-resolution 3D imaging method using chirped optical frequency combs based on convolution analysis of the spectral interference fringe,” OSA Continuum 3(1), 20–30 (2020). [CrossRef]
17. Y. Wang, G. Xu, S. Xiong, and G. Wu, “Large-field step-structure surface measurement using a femtosecond laser,” Opt. Express 28(15), 22946–22961 (2020). [CrossRef]
18. Q. D. Pham and Y. Hayasaki, “Combining phase images measured in the radio frequency and the optical frequency ranges,” Opt. Lett. 42(11), 2062–2065 (2017). [CrossRef]
19. K. Shibuya, T. Minamikawa, Y. Mizutani, H. Yamamoto, K. Minoshima, T. Yasui, and T. Iwata, “Scan-less hyperspectral dual-comb single-pixel-imaging in both amplitude and phase,” Opt. Express 25(18), 21947–21957 (2017). [CrossRef]
20. Q. D. Pham and Y. Hayasaki, “Optical frequency comb profilometry with a compressive sensing-based single-pixel camera composed of digital micromirror devices and a two-frequency method for meter-order depth measurements,” J. Micro/Nanolith. MEMS MOEMS 14(4), 041305 (2015). [CrossRef]
21. D. Kim, Y. Lu, J. Park, B. Kim, L. Yan, L. Yu, K.-N. Joo, and S.-W. Kim, “Rigorous single pulse imaging for ultrafast interferometric observation,” Opt. Express 27(14), 19758–19767 (2019). [CrossRef]
22. K. Koerner, G. Pedrini, I. Alexeenko, T. Steinmetz, R. Holzwarth, and W. Osten, “Short temporal coherence digital holography with a femtosecond frequency comb laser for multi-level optical sectioning,” Opt. Express 20(7), 7237–7242 (2012). [CrossRef]
23. J. S. Oh and S. W. Kim, “Femtosecond laser pulses for surface-profile metrology,” Opt. Lett. 30(19), 2650–2652 (2005). [CrossRef]
24. H. Zhao, Z. Zhang, X. Xu, H. Zhang, J. Zhai, and H. Wu, “Three-Dimensional Imaging by Frequency-Comb Spectral Interferometry,” Sensors 20(6), 1743 (2020). [CrossRef]
25. Y. Na, C.-G. Jeon, C. Ahn, M. Hyun, D. Kwon, J. Shin, and J. Kim, “Ultrafast, sub-nanometre-precision and multifunctional time-of-flight detection,” Nat. Photonics 14(6), 355–360 (2020). [CrossRef]
26. Q. D. Pham and Y. Hayasaki, “Optical frequency comb profilometry using a single-pixel camera composed of digital micromirror devices,” Appl. Opt. 54(1), A39–44 (2015). [CrossRef]
27. W. K. Pratt, J. Kane, and H. C. Andrews, “Hadamard transform image coding,” Proc. IEEE 57(1), 58–68 (1969). [CrossRef]
28. M. F. Duarte, M. A. Davenport, D. Takhar, J. N. Laska, T. Sun, K. F. Kelly, and R. G. Baraniuk, “Single-pixel imaging via compressive sampling,” IEEE Signal Process. Mag. 25(2), 83–91 (2008). [CrossRef]
29. Q. D. Pham and Y. Hayasaki, “Optical frequency comb interference profilometry using compressive sensing,” Opt. Express 21(16), 19003–19011 (2013). [CrossRef]
30. M. J. Sun, L. T. Meng, M. P. Edgar, M. J. Padgett, and N. Radwell, “A Russian Dolls ordering of the Hadamard basis for compressive single-pixel imaging,” Sci. Rep. 7(1), 3464 (2017). [CrossRef]
31. M. P. Edgar, G. M. Gibson, R. W. Bowman, B. Sun, N. Radwell, K. J. Mitchell, S. S. Welsh, and M. J. Padgett, “Simultaneous real-time visible and infrared video with single-pixel detectors,” Sci. Rep. 5(1), 10669 (2015). [CrossRef]
32. M.-J. Sun, M. P. Edgar, G. M. Gibson, B. Sun, N. Radwell, R. Lamb, and M. J. Padgett, “Single-pixel three-dimensional imaging with time-based depth resolution,” Nat. Commun. 7(1), 12010 (2016). [CrossRef]
33. M. Li, L. Yan, R. Yang, and Y. Liu, “Fast single-pixel imaging based on optimized reordering Hadamard basis,” Acta Phys. Sin. 68(6), 064202 (2019). [CrossRef]
34. M. F. Li, Y. R. Zhang, X. F. Liu, X. R. Yao, K. H. Luo, H. Fan, and L. A. Wu, “A double-threshold technique for fast time-correspondence imaging,” Appl. Phys. Lett. 103(21), 211119 (2013). [CrossRef]
35. J. Lee, S. Han, K. Lee, E. Bae, S. Kim, S. Lee, S. W. Kim, and Y. J. Kim, “Absolute distance measurement by dual-comb interferometry with adjustable synthetic wavelength,” Meas. Sci. Technol. 24, 045201 (2013). [CrossRef]
36. Y.-S. Jang and S.-W. Kim, “Distance Measurements Using Mode-Locked Lasers: A Review,” Nanomanuf. Metrol. 1(3), 131–147 (2018). [CrossRef]
37. S. Xiong, Y. Wang, Y. Cai, J. Liu, J. Liu, and G. Wu, “Calculating the Effective Center Wavelength for Heterodyne Interferometry of an Optical Frequency Comb,” Appl. Sci. 8(12), 2465 (2018). [CrossRef]
38. M. Harwit, Hadamard transform optics (Elsevier Science, 2012).