Questions: Perfect Reconstruction Filter Banks and Constraints

Question 1

In a two-band filter bank, a designer uses a perfect lowpass and perfect highpass filter (ideal brickwall filters with no frequency overlap). Why does this NOT achieve perfect reconstruction?

• A. Brickwall filters introduce too much delay, violating the distortion-free condition
• B. Non-overlapping brickwall filters cannot satisfy the power complementary condition required for aliasing cancellation in the synthesis stage
• C. The downsampling operation requires overlapping filters to avoid spectral gaps
• D. Brickwall filters are FIR, and PR requires IIR synthesis filters

Answer: (B)

Perfect reconstruction requires two conditions: aliasing cancellation (A(z) = 0) and distortion-free reconstruction (T(z) = cz^{-n}). The aliasing cancellation condition requires the analysis and synthesis filters to satisfy a power complementary relationship: |H_0(e^{jω})|² + |H_0(e^{j(ω−π)})|² = 1. Ideal brickwall filters do not satisfy this — their transition from 1 to 0 is a step function, not a smoothly complementary shape. Non-overlapping filters eliminate aliasing differently (by preventing it from occurring), but their infinitely sharp transition bands are physically unrealizable and cannot satisfy the algebraic PR conditions.

Question 2

Perfect reconstruction in a filter bank is a property of the individual analysis filters alone — if each analysis filter is well-designed, reconstruction will be exact.

• A. True, because good frequency selectivity in analysis ensures no information is lost
• B. False — PR is a property of the analysis-synthesis pair together; the synthesis filters must be specifically designed to cancel the aliasing introduced by the analysis filters and downsampling
• C. True, because the synthesis filters are just the inverses of the analysis filters by construction
• D. False — PR depends on the downsampling factor, not the filter design

Answer: (B)

PR is not a property of individual filters — it is a property of the complete analysis-downsample-upsample-synthesis cascade. The downsampling operation introduces aliasing (frequency folding), and the synthesis filters must be specifically designed relative to the analysis filters so that this aliasing cancels exactly when the subbands are recombined. The conjugate quadrature filter (CQF) solution defines synthesis filters as specific transformations of the analysis prototype precisely to guarantee this cancellation. Designing analysis filters independently without constraining the synthesis filters provides no PR guarantee.

Question 3

Perfect reconstruction filter banks must use overlapping filters in the frequency domain — non-overlapping (brickwall) filters cannot satisfy PR constraints.

• T. True
• F. False

Answer: True

This is a key counterintuitive result. One might expect that filters with perfectly separated frequency bands (no overlap) would be ideal since downsampling of non-overlapping bands introduces no aliasing. However, ideal brickwall filters are unrealizable in practice, and more importantly, the PR condition requires the power complementary relationship |H_0(e^{jω})|² + |H_0(e^{j(ω−π)})|² = 1, which a step-function brickwall filter cannot satisfy. Practical PR filter banks use carefully shaped overlapping filters where the transition bands are designed so that aliasing from neighboring subbands cancels algebraically in the synthesis stage.

Question 4

Perfect reconstruction means that each individual subband output of the analysis filter bank has no aliasing — each subband is a clean, alias-free version of its spectral portion.

• T. True
• F. False

Answer: False

This is a common misconception. Downsampling after the analysis filters does introduce aliasing into each subband — frequency components fold back on themselves. PR does not mean each subband is alias-free; it means the aliasing introduced in all subbands cancels exactly when they are synthesized (upsampled and recombined). This algebraic cancellation across subbands is the core mechanism. A 'no-aliasing' bank would require non-overlapping brickwall filters to prevent aliasing in the first place, which is a completely different (and less practical) design philosophy than PR.

Question 5

What are the two conditions that must be simultaneously satisfied for a two-band filter bank to achieve perfect reconstruction, and why must both hold?

Model answer: The two conditions are: (1) aliasing cancellation — the term A(z)X(−z) in the reconstructed signal must be zero, meaning the synthesis filters are chosen so that aliased components from H_0 and H_1 cancel exactly when summed; and (2) distortion-free reconstruction — the remaining term T(z) must equal a pure gain and delay (cz^{-n}), meaning the signal passes through without amplitude or phase distortion beyond a constant delay. Both must hold because they address different failure modes: aliasing cancellation ensures that frequency-folded artifacts from downsampling don't contaminate the output, while the distortion condition ensures the signal's spectral shape is preserved. Satisfying aliasing cancellation alone still allows amplitude and phase distortion; satisfying distortion-free alone still allows aliasing artifacts.

The z-transform analysis of the two-band filter bank yields X̂(z) = T(z)X(z) + A(z)X(−z). For X̂(z) = cz^{-n}X(z) (perfect reconstruction with a delay), we need A(z) = 0 (kill the aliasing term) and T(z) = cz^{-n} (make the distortion term a pure delay). The conjugate quadrature filter family achieves both simultaneously through a specific relationship between the analysis and synthesis filter prototypes.