Hybrid Time-Shifted Feature Augmentation for Compact Photonic Time-Delay Reservoir Computing

Photonic time-delay reservoir computing (TDRC) exploits the nonlinear dynamics of a delayed optical component to emulate a high-dimensional reservoir and offers a compact hardware platform for temporal information processing. Its speed, however, is limited by temporal multiplexing, which updates virtual nodes sequentially and creates a direct trade-off between reservoir dimensionality and processing rate. Building on time shifted reservoir state processing and delayed input augmentation, this work investigates a hybrid time shifted feature augmentation method (HTSRC) for compact photonic TDRC. By combining delayed input encoding with exponentially parameterized delayed reservoir state features, HTSRC enriches the effective readout representation beyond conventional time multiplexing. Accurate prediction can therefore be achieved with fewer physically generated virtual nodes in the tested tasks, while the final readout state matrix is expanded digitally. The feasibility of the proposed compact mechanism is further supported by experimental results from a lightweight photonic hardware implementation.