Photonic crystal filters can greatly increase the transmission capacity of optical communication systems, which is a research hotspot for optical wavelength division multiplexing technology. Tunable optical filters are key components in optical wavelength division multiplexing systems. A monotonic functional relationship can easily be formed between the transmission peak wavelength of a single channel filter and tuning parameters to help achieve precise tuning. However, the information capacity of this filter is limited. The coupling relationship between dual-channel or even multi-channel filtering structures and transmission peak wavelength is complex. The overall translation of multi-channel transmission peaks is mainly focused on current research and lacks discussion on precise tuning. We design a tunable dual-channel filter based on BaTiO₃ (BTO) defect layers and introduce a genetic algorithm to achieve precise tuning of dual channels. This one-dimensional photonic crystal filter changes the refractive index of BTO by altering applied voltages, thereby achieving separate tuning of dual transmission peak positions. Purposeful genetic algorithms can both automatically find peaks and ensure transmissivity, relying on computer computing power to realize automatic tuning of filtering wavelengths.

By adopting the transfer matrix theory, transmission spectra of the filter are obtained, and surface functions of two transmission peaks with the variation of two voltages are fitted. By combining surface projection with surface functions, precise tuning of the positions of two transmission peaks can be achieved. Meanwhile, a genetic algorithm is also introduced to achieve automatic tuning. By employing an algorithm framework based on purposes, the program can automatically find voltage combinations that match target peaks, thus avoiding manual involvement in analyzing surfaces and functions. Additionally, the tuning process can be simplified, and efficiency can be improved with the help of computing power.

Structural parameters of the crystal filter are analyzed. Fig. 3 shows that when the period number is set to be 4, the requirements of wavelength division multiplexing can be met, and the overall size of the structure is also reduced. Meanwhile, when the voltage difference between two external voltages is less than 50 V, a transmissivity over 78% can be ensured to satisfy the filtering requirements (Fig. 4). By changing the thickness of the air layer in the middle of the filter, filtering can also be switched between bands such as (S, L), (E, C), and (C, L) (Fig. 5). By conducting surface fitting, we find that the tunable ranges of the two transmission peaks are 1499-1533 nm and 1572-1615 nm respectively. The specific relationship between the position wavelengths of two transmission peaks and the voltages are shown in Eqs. (8) and (9). Traditional methods require precise tuning of the positions of two transmission peaks by curved surface projection. We take a transmission peak fixed at 1510 nm as an example and obtain a tunable function for the other transmission peak [Eq. (11)]. By combining Eq. (10), the corresponding external voltage combination can be obtained. In selecting an appropriate fitness function [Eq. (13)] with a crossover rate of 0.6 and a mutation rate of 0.4, the genetic algorithm can autonomously search for external voltage combinations that match the target transmission peaks about an average of 400 generations. Additionally, we test the convergence performance of genetic algorithms for tuning three target wavelengths, including (1510 nm, 1588 nm), (1515 nm, 1595 nm), and (1520 nm, 1599 nm). Fig. 9 shows that when the target wavelengths are (1510 nm, 1588 nm), the transmissivity of both peaks is higher than 94% with a convergence generation of 416 generations. Fig. 10 shows that when the target wavelengths are (1515 nm, 1595 nm) and (1520 nm, 1599 nm), the transmissivity is higher than 92% and 97% respectively, with convergence generations of 373 and 395 generations. Due to the avoidance of relatively complex manual analysis processes, genetic algorithms can improve tuning efficiency with the help of computer computing power.

We design a photonic crystal dual-channel filter based on BTO defect layer. A genetic algorithm is introduced into the filter tuning for efficient and fast dual-channel simultaneous accurate filtering. The convergence performance of the genetic algorithm for tuning three target wavelengths is tested, including (1510 nm, 1588 nm), (1515 nm, 1595 nm), and (1520 nm, 1599 nm). The voltage combinations for the above target wavelengths with transmissivity greater than 90% are obtained, and their convergence generations are 416, 373, and 395 respectively. Similar calculations also indicate that genetic algorithms can converge within an average of 400 generations. By comparing and validating the above data using distribution surfaces and fitting tuning functions, we demonstrate the accuracy and efficiency of genetic algorithms in tuning. Meanwhile, the designed filter can also achieve dual-channel step tuning in the (S, L), (E, C), and (C, L) bands by changing the thickness of the intermediate air layer. Finally, our study will provide references for the design and application of photonic crystal filters.