Fast Spectrogram-Based Method for Identifying Relative SNR Variations in Narrowband Signals

Abstract

This paper proposes a fast, agile, low-complexity method for tracking relative signal-to-noise ratio (SNR) variations directly from spectrogram representations of narrowband signals. The approach operates on complex spectrogram values and employs simple statistical estimators, namely the maximum and median across frequency bins, to estimate a peak to background contrast metric within each time window. The proposed technique does not require access to the original time-domain signal, making it suitable for applications where only time–frequency representations are available. The method is evaluated using synthetic sinusoidal and chirp signals corrupted by additive white Gaussian noise over a wide SNR range (0–100 dB). Experimental results show a strong linear relationship between the estimated SNR-related metric, expressed in dB, and the reference SNR (namely in the 10-60 dB range), with correlation coefficients close to unity and low variability across different signal parameters. The proposed metric should be interpreted as a relative spectrogram SNR indicator rather than a calibrated SNR estimate, making it particularly suitable for low complexity applications such as interference/spectrum monitoring and signal quality tracking in systems where only spectrogram data is available or where computational simplicity is important.