Questions: Wavelet Transform and Multiresolution Analysis

Question 1

An audio signal contains both a 1 ms mechanical click and a 3 Hz structural vibration. An analyst uses STFT with a fixed window. What fundamental problem arises, and how do wavelets address it?

• A. STFT handles both equally well; wavelets offer no advantage for mixed-frequency signals
• B. STFT forces a single window size, so the analyst must choose between time resolution (good for the click) and frequency resolution (good for the vibration) — wavelets adaptively scale, using short durations for the high-frequency click and long durations for the slow vibration
• C. Wavelets cannot resolve low-frequency content below 10 Hz, so STFT is required for the vibration component
• D. STFT provides better time localization than wavelets for transient events like the click

Answer: (B)

This is the fundamental limitation that wavelets solve. STFT applies a fixed window everywhere — short windows give good time resolution but smear frequency content, long windows give good frequency resolution but blur events in time. You cannot have both at once. Wavelets abandon the fixed window: a compressed mother wavelet captures the click with fine time resolution, while a stretched copy of the same wavelet resolves the slow vibration with fine frequency resolution. This automatic adaptation — short time/coarse frequency at high frequencies, long time/fine frequency at low frequencies — matches the natural structure of many real signals.

Question 2

In the discrete wavelet transform, highpass filtering produces 'detail coefficients' and lowpass filtering produces 'approximation coefficients.' What do these represent?

• A. Detail coefficients capture slow background variations; approximation coefficients capture rapid transient features
• B. Detail coefficients capture fast variation (high-frequency content) from the highpass filter; approximation coefficients capture slow variation (low-frequency content) from the lowpass filter
• C. Both capture the same information; the distinction is only about computational efficiency
• D. Approximation coefficients are redundant and are discarded once detail coefficients are computed at all levels

Answer: (B)

The names are intuitive: the approximation is the coarse, smooth version of the signal, while the details are the fine-scale variations that distinguish the approximation from the original. The lowpass filter passes low-frequency content → approximation coefficients. The highpass filter passes high-frequency content → detail coefficients. At each DWT level, the approximation is recursively decomposed into a coarser approximation and another layer of details, building up the multi-resolution structure. The detail coefficients at each level capture features at a specific scale, and together with the final approximation they provide a complete reconstruction of the original signal.

Question 3

The Fourier transform provides better time-frequency localization than the wavelet transform for analyzing non-stationary signals.

• T. True
• F. False

Answer: False

The Fourier transform provides no time localization at all — it tells you which frequencies are present in the entire signal but cannot tell you when they occur. Wavelets are specifically designed to provide time-frequency localization: by correlating the signal with scaled and shifted copies of a localized mother wavelet, the CWT locates both when and at what frequency events occur. For non-stationary signals (where frequency content changes over time), wavelets are strictly superior to Fourier methods in this regard. The Fourier transform is best for stationary signals where time-localization is unnecessary.

Question 4

A compressed (scaled-down) wavelet has a higher oscillation frequency and provides finer time resolution than a stretched version of the same mother wavelet.

• T. True
• F. False

Answer: True

Scaling is the core mechanism of wavelets. When a mother wavelet ψ is compressed (small scale parameter), it oscillates faster — its temporal support is narrow, giving fine time resolution but spanning a broader frequency band. When stretched (large scale parameter), it oscillates slowly — its temporal support is wide, giving fine frequency resolution but poor time localization. This trade-off is not a fixed limitation (as in STFT) but an adaptive feature: the wavelet transform evaluates the signal at every scale, so both high-frequency events (captured by compressed wavelets) and low-frequency structure (captured by stretched wavelets) are analyzed optimally.

Question 5

Explain why wavelets are particularly well-suited for signal compression compared to the STFT or standard Fourier transform.

Model answer: Wavelets concentrate signal energy into a small number of large coefficients by exploiting the structure of most real signals: smooth regions produce small detail coefficients, while sharp features and discontinuities produce large detail coefficients localized in time. Compression works by thresholding small coefficients to zero — removing them causes little reconstruction error. The DWT also produces a non-redundant decomposition (exactly as many coefficients as input samples), making storage efficient. In contrast, the Fourier transform spreads energy of localized features (like discontinuities) across many coefficients, and the STFT is redundant and uses a fixed resolution that may not match the signal's structure. Wavelets match resolution to signal content, which is why JPEG 2000 and audio codecs use them.

The key is sparsity: compression works best when most coefficients are near zero. Wavelets achieve sparsity for natural signals because most of the content is in smooth regions (few detail coefficients needed) with a few sharp transitions (localized large coefficients). The Fourier transform's coefficients are not sparse for signals with discontinuities — a single edge requires many sinusoids to represent, spreading energy widely.