Distributed estimation of statistical parameters in an experimental network of single-transistor chaotic sensor nodes using frequency spectra and geometrical reconstruction

The possibility of performing useful computations on hardware substrates different from digital computers, such as networks of coupled nonlinear oscillators, represents an important perspective for increasing the pervasiveness of intelligent systems. Recent simulations have suggested that analog electronic circuits as simple as single-transistor chaotic oscillators could aggregate sensor readings over a wireless communication channel, offering direct readout of statistical parameters such as mean and variance without the need to interrogate individual nodes. In this paper, we present a comprehensive experimental demonstration of the viability of this concept using a benchtop apparatus comprising a complex network of coupled chaotic oscillators, each one controlled by a hypothetical sensor. By recording the ensemble-level activity corresponding to signals recorded by far-field and near-field antennas, it is shown that accurate estimation of the mean and variance is possible across a wide range of operating conditions. Modulating the coupling strength globally reveals transitions between distinct regimes, characterized by diverse collective responses to the distribution of sensor readings, including varying levels of coherent activity summation. Moreover, it is demonstrated that the statistical parameters of interest can be estimated from two independent representations of the signals: frequency spectra and time-lag reconstructions of the underlying geometry. Altogether, these results affirm the physical feasibility of distributed sensing in a network of elementary chaotic oscillators, paving the way to field experiments and eventually applications.