Physics
Time-Frequency Analysis
Quick fact
A standard Fourier transform shows all the frequencies present in a signal, but it loses all information about when those frequencies occur. Time-frequency analysis fixes this by letting you watch the spectrum evolve moment by moment.
Why this is interesting
You can easily see when a sound is loud or quiet on a waveform, but how do you see which note is being played at each moment—like reading a musical score from a recording?
Read the full explanation
Understanding Time-Frequency Analysis
Imagine listening to a song. Your ears simultaneously perceive the melody (changes over time) and the harmonies (frequencies). A simple graph of amplitude over time (waveform) shows when sounds start and stop but not their pitch. Conversely, a frequency plot (spectrum) shows all the notes played during the whole song, but not in what order. Time-frequency analysis combines both: it chops the signal into short overlapping chunks, computes the spectrum of each chunk, and stacks them side by side. The result is a spectrogram—a 2D plot where time runs left to right, frequency runs bottom to top, and color represents intensity. This lets you see precisely when a guitar strums a chord or when a bird chirps its varying pitch.
A deeper explanation
The core mechanism is the short-time Fourier transform (STFT), which multiplies the signal by a window function (like a brief pulse) and applies the Fourier transform to that segment. Sliding the window along the signal yields a series of spectra. However, there is a fundamental trade-off: shorter windows give better time resolution (you see rapid changes) but poorer frequency resolution (you can't distinguish close frequencies), while longer windows improve frequency resolution at the cost of blurring events in time. This is the Heisenberg-Gabor uncertainty principle of signal analysis. To overcome this limit, wavelet transforms use variable-length windows—short for high frequencies, long for low frequencies—giving a more natural multi-resolution analysis. Time-frequency analysis matters because most real-world signals, from speech to electrocardiograms, are non-stationary: their properties change over time. Without this tool, we would miss the dynamic structure of these signals.