Analytic-signal-based input-output modeling inspires passband signal learning for spin-wave reservoir computing

Spin-wave reservoir computing (RC) utilizes spatiotemporal dynamics of spin waves in a magnetic thin film as a physical reservoir to realize effective neuromorphic computation. Because of the complicated nature of spin-wave dynamics, the spin-wave RC systems exhibit intrinsic properties that are potentially different from software-based RC systems, such as echo-state networks, thus making these two systems suitable for different application contexts. Therefore, a deeper understanding of spin-wave dynamics and their role in neuromorphic computing systems is crucial for more effective system design and practical applications. Here, we construct an analytic-signal-based input-output model for spin-wave RC systems. The proposed model allows for the direct construction of system responses without the need for physical experiments or numerical simulations, given knowledge of the system's standard responses under low input power. Inspired by our model, we refine the spin-wave RC system by proposing an output block based on passband signal learning to exploit the wave-based characteristics of the system responses. We evaluate passband signal learning in the context of multipath interference (MPI) mitigation. Our results demonstrate that spin-wave RC systems with passband signal learning can more effectively solve the MPI mitigation problem, owing to similar inherent properties between the spin-wave RC systems and the communication channels. We expect this work to provide deep insights into spin-wave physics and similar spin-wave-based neuromorphic computing systems.