There is an increasing need for everyday communications to be both secure and resistant to external perturbations. We have therefore created an experimental implementation of the coupling-function-based secure communication protocol, in order to assess its robustness to channel noise. The transmitter and receiver are implemented on single-board computers, thereby facilitating communication of the analog electronic signals. The information signals are encrypted at the transmitter as the timevariability of nonlinear coupling functions between dynamical systems. This results in a complicated nonlinear mixing and scrambling of the information. To replicate the channel noise, analog white noise is added to the encrypted signals. After digitization at the receiver, the decryption is performed with dynamical Bayesian inference to take account of time-varying dynamics in the presence of noise. The dynamical Bayesian approach effectively separates the deterministic information signals from the perturbations of dynamical channel noise. The experimental realization has demonstrated the feasibility, and established the performance, of the protocol for secure, reliable, communication even with high levels of channel noise.