Attentive signal processing for detection of sinusoids in noise

The paper presents a non-linear signal processing system that is suitable for detection and tracking individual, possibly non-stationary, components of the complex signal. The core signal processing algorithm is inspired by the ways human auditory nerve responds to tonal and noise stimuli. The performed experiments confirm ability of the system to lock its attention onto the sinusoidal signals embedded in the noise with low SNR and produce accurate estimates of sinusoid frequencies. Precision of the estimates is tested with Monte-Carlo trials over the entire space of possible amplitudes, frequencies and initial phases. The amplitude has spanned a reasonably wide dynamic range of 100 dB.