The wavemeter is one of the most significant tools for identifying the wavelength of light, which holds broad applications in remote sensing, spectral calibration, fluorescence spectroscopy, atomic physics, and high-precision metrology. Traditionally, there are three categories according to the strategies applied: dispersive optics, Fourier transform systems, and narrowband filters. Wavemeters based on dispersive optics split light into spatially separated detectors and the resolution depends on the dispersive grating, the size of detectors, and the distance between them. Fourier transform wavemeters relying on Fabry-Perot cavities can access higher accuracy in wavelength recognition compared with those wavemeters based on dispersive optics. Wavemeters using narrowband filters selectively transmit light with designated wavelengths.

The trade-off relationship between resolution and footprint always exists in a wavemeter. Therefore, it is essential to solve this problem based on a new wavelength-distinguishing principle. In the last decade, wavemeters and spectrometers based on scattering speckle have emerged as an important method for balancing the trade-off between resolution and footprint. However, these studies used infrared-sensitive detectors when the wavelengths of light reach the infrared range. The commonly used infrared detecting instrument is based on integrated InGaAs detectors, which are expensive and lack high-performance sensitivity. Therefore, finding a way to overcome the above problems is highly desirable.

Here, we propose a near-infrared (NIR) speckle wavemeter based on nonlinear frequency conversion. In our experiment as shown in Fig. 1(a), we use a periodically poled lithium niobate (PPLN) waveguide to realize the frequency up-conversion of the infrared light. The measured normalized FD efficiency and the quadratic scaling of the nonlinear intensity with the pump power at 1550 nm center wavelength with the temperature controlled at 34 ℃ is shown in Fig. 1(b) and 1(c), respectively. The generated nonlinear signal in the visible waveband is incident on titanium dioxide (TiO2) powder to generate scattering speckles, which are captured by a silicon-based charge-coupled device (CCD). Silicon-based visible light detectors are much cheaper and easier to manufacture. The deep learning method is used to map the relation between visible light speckles and NIR wavelengths, as shown in Fig. 1(d).

Fig. 1. (a) Experiment setup of NIR speckle wavemeter based on nonlinear frequency conversion; (b) FD efficiency of the PPLN waveguide used in the experiment at 34 ℃ with sinc2 fit; (c) The quadratic scaling of the nonlinear intensity with the pump power; (d) Architecture of deep learning neuron network ResNet50 used in NIR wavemeter.

To verify the resolution of the wavemeter, we systematically adjusted the NIR wavelengths from 1549.800 nm with an interval of 1 pm up to 1549.806 nm. Part of the collected speckle patterns of different incident NIR wavelengths is illustrated in Fig. 2(a).  After the training process, the test dataset is used to test the performance of our trained model. Figure 2(b) shows that the wavelength resolution accuracy is increasing with the training epochs, which begins to converge quickly at the 100th epoch. The normalized confusion matrix for wavelength recognition from 1549.800 nm to 1549.806 nm shows the good performance of our neural network in the wavelength recognition task as illustrated in Fig. 2(c). We demonstrate that our nonlinear NIR wavemeter can easily reach a resolution of 1 pm in the circumstance of frequency-doubling (FD). We also analyze the correlation of speckles between different wavelengths and the noise level of the system. Our wavemeter can also provide power information of the laser. Besides, we further demonstrate the recognition ability of dual-wavelength situations in our system. In summary, our proposed NIR speckle wavemeter based on nonlinear frequency conversion provides an effective solution for high-resolution, low-cost wavelength measurements, and exhibits significant potential across various application domains. The method proposed here might even be applied to other non-visible range, such as ultra-violet and far-infrared spectrum. This study has extended the application scope of wavemeter technology, offering broader possibilities for spectral analysis.

Fig. 2. Single frequency-doubling wavelength recognition analysis. (a) Visible light speckle patterns corresponding to different NIR light; (b) The accuracy increases with the epochs at the condition of 1 pm wavelength interval from 1549.800 nm to 1549.806 nm; (c) The confusion matrix of NIR wavelength detection with 50 images in each test dataset.

This research is published in “Yiwei Sun, Fengchao Ni, Yiwen Huang, Haigang Liu, and Xianfeng Chen, Near-infrared speckle wavemeter based on nonlinear frequency conversion, Optics Letters, 48(15), 4049-4052 (2023)”.

Link: https://opg.optica.org/ol/fulltext.cfm?uri=ol-48-15-4049&id=535472