Questions: Convolution Theorem and Frequency Domain Applications

Question 1

You want to convolve a 10-second audio signal (441,000 samples at 44,100 Hz) with a 1-second room impulse response (44,100 samples) using the convolution theorem. What is the computational advantage over direct time-domain convolution?

• A. The FFT approach is O(N²) while direct convolution is O(N log N) — FFT is more accurate but slower
• B. The FFT approach reduces the cost from O(N²) to O(N log N) via the FFT algorithm — a speedup factor of roughly N/log N
• C. Both approaches have the same complexity; the only difference is numerical precision
• D. The FFT approach only helps for very short filters — 1-second impulse responses still require direct convolution

Answer: (B)

Direct convolution of two N-sample signals requires O(N²) multiplications. Using the convolution theorem: compute two FFTs (O(N log N) each), multiply spectra pointwise (O(N)), then compute the inverse FFT (O(N log N)) — total O(N log N). For N = 44,100, this is a speedup factor of roughly N/log N ≈ 2,800. Longer impulse responses make this advantage even larger. Real-time audio reverb (2–3 second impulse responses) is computationally feasible only because of FFT convolution.

Question 2

A system has frequency response H(f) with |H(f)| = 1 for |f| < 1,000 Hz and |H(f)| = 0 for |f| > 1,000 Hz. What does Y(f) = X(f)H(f) tell you about the output signal y(t)?

• A. The output y(t) is the time-reversed version of x(t), since high frequencies are removed
• B. Frequency components of x(t) above 1,000 Hz are eliminated; components below are passed through unchanged
• C. The output is y(t) = x(t) + h(t) — the system adds the impulse response to the input
• D. High-frequency components are slowed down in time rather than removed

Answer: (B)

The convolution theorem says Y(f) = X(f)·H(f): each frequency component of X(f) is multiplied by H(f)'s value at that frequency. Where |H(f)| = 1, the component passes unchanged; where |H(f)| = 0, it is nullified. This is a perfect low-pass filter — all energy above 1,000 Hz is removed, all energy below is preserved. This pointwise multiplication is why filter design is natural in the frequency domain: you specify what to keep by defining H(f), and the convolution theorem guarantees the time-domain filter achieves exactly that effect.

Question 3

The convolution theorem states that convolution in the time domain corresponds to multiplication in the frequency domain.

• T. True
• F. False

Answer: True

This is the precise statement of the theorem: if y(t) = x(t) * h(t) in the time domain, then Y(f) = X(f)·H(f) in the frequency domain. The dual also holds: multiplication in time corresponds to convolution in frequency. The time-to-frequency direction is most useful for signal processing because it replaces the computationally expensive integral of convolution with cheap pointwise multiplication of spectra, and because frequency-domain thinking makes filter design intuitive.

Question 4

The convolution theorem and frequency-domain filtering apply to any system, including nonlinear ones, as long as the input and output are both bounded signals.

• T. True
• F. False

Answer: False

The convolution theorem's application to system analysis requires linearity and time-invariance (LTI). For an LTI system, the output is y(t) = x(t) * h(t), and taking the Fourier transform gives Y(f) = X(f)H(f). For nonlinear systems, the output cannot be written as a convolution with a fixed impulse response — the system's behavior depends on input amplitude, creating cross-frequency interactions (harmonic distortion, intermodulation) that a single H(f) cannot capture. Bounded signals are not the relevant condition; LTI structure is.

Question 5

Explain what H(f) = ℱ{h(t)} represents physically, and why it is called the 'frequency response' of a system.

Model answer: H(f) describes how the system treats each individual frequency component: if you input a pure sinusoid at frequency f₀, the output is a sinusoid at the same frequency with amplitude scaled by |H(f₀)| and phase shifted by the argument of H(f₀). Since any input can be decomposed into sinusoids via the Fourier transform, knowing H(f) at every frequency completely characterizes the system — each frequency is handled independently, then the results superimpose.

This is why filter design is done in the frequency domain. You want a low-pass filter: set |H(f)| ≈ 1 for low f and ≈ 0 for high f. You want a notch at 60 Hz: set H(60) ≈ 0. The convolution theorem translates this frequency-domain specification directly into time-domain behavior: applying the filter to any input in the time domain produces exactly the frequency shaping you specified. The impulse response h(t) and frequency response H(f) are two equivalent, complementary descriptions of the same LTI system.