Objective  A tunable optical filter is the key component in the optical communication systems and optical processing systems. By tuning the central wavelength, it can be used to choose a signal with arbitrary wavelength according to the practical requirement. However, due to the high loss and large size, optical filter based on discrete devices cannot meet the requirements of some photonic signal processor. Recently, based on Mach-Zehnder interferometers network and ring-assisted Mach-Zehnder interferometer, some research groups have proposed integrated optical filter. However, this type of optical filter has the disadvantages of being difficult to train and having a single waveform, which restricts its application in the fields of multi-purpose adaptive signal processing. For improving the flexibility of optical filter, a novel tunable optical filter with an integrated photonic reservoir computing (RC) is proposed. Since the filtering properties can be controlled by intelligence algorithm, this optical filter, which improves the flexibility in applications, can be widely applied in optical cross interconnection system and microwave photon signal shaping.

  Methods  Firstly, the structure of integrated photonic RC is constructed in detail, and scattering matrix theory is used to analyze the transmission function of integrated photonic RC. Then, the simulated transmission spectra of the reservoir are carried out by simulation software. Particle swarm optimization (PSO) algorithm is matched for training reservoir transmission spectra in optical domain to find optimal weights. Based on thermo-optical effects, optical weights are implemented by optical modulators (OMs). During training the weights in readout layer, OMs are used to adjust the amplitudes and phases of the optical signal. Using this integrated photonic RC chip, the infinite (IIR) and finite (FIR) impulse response optical filters are realized. Finally, by adjusting the parameters of waveguide (WG) and directional coupler (DC) in the reservoir, the filtering properties is studied.

  Results and Discussions   The achieved IIR and FIR optical filter waveform are almost exactly matched to the ideal waveform (Fig.5(a), Fig.6(a)). The error value of the training results decreases in a step-like trend with the increase of the number of iterations, and eventually tends to be stable (Fig.5(b), Fig.6(b)). Based on the IIR optical filter simulation results, the effect of the free spectral range (FSR) on the WG length is analyzed (Fig.8). The WG length is negatively correlated with the FSR. As the WG length increases, the corresponding FSR becomes smaller. In addition, the influence of the DC splitting ratio on the transmission power is analyzed (Fig.9). According to the actual needs, the adjustment of different filtering intensities is achieved by setting the splitting ratio of the even number of output nodes of the reservoir. Moreover, the filtering wavelength, which is influenced by the phase of the WG in the reservoir from 0 to 3/2π, is continuously adjustable in the FSR of 1.18 nm (Fig.10).

  Conclusions  In this study, a novel tunable optical filter basd on 12-node plum shaped integrated photonic RC chip is constructed. The PSO algorithm is used for training photonic RC weights to realize the IIR and FIR optical filters. The control of FSR is achieved by adjusting the length of the waveguide in the reservoir. Under the constant filter waveform, the filtering wavelength can be continuously tuned in the FSR by adjusting the phase of the WG in the reservoir (0-3π/2). The feasibility of this optical filter is verified by theory and simulation, and its tailorable performance can be used in multi-purpose adaptive signal processing application.