TL;DR
This paper proves that certain standard methods for reconstructing bandlimited signals from their samples can diverge significantly, even in the limit. The authors demonstrate that these reconstruction procedures don't just exhibit occasional divergence (lim sup), but can diverge strongly, meaning the error grows unboundedly. This has important implications for the reliability of signal reconstruction techniques.
What Problem Does This Paper Address?
Sampling theory is concerned with reconstructing a signal from its discrete samples. For bandlimited signals (those whose frequency content is limited), the Paley-Wiener space ($PW_\pi^1$) and Hardy space ($H^1$) are key function classes. While the Shannon sampling theorem guarantees perfect reconstruction under ideal conditions, practical reconstruction methods often involve approximations and infinite series. A crucial question in this field is the convergence behavior of these reconstruction procedures. Previous work showed that some methods could exhibit "lim sup divergence," meaning the error could grow arbitrarily large for certain subsequences of parameters. However, it was unclear if this divergence was inherent in the limit or only occurred for specific, pathological subsequences. This paper tackles the stronger question: do these reconstruction methods diverge fundamentally, even in the limit, for some signals?
Methods
The authors investigate three canonical reconstruction methods for $PW_\pi^1$ and one for $H^1$. Their core approach involves constructing specific functions within these spaces that exhibit the desired divergence behavior. For $PW_\pi^1$, they demonstrate "strong divergence" by showing that the reconstruction error tends to infinity as the number of samples considered increases. This is a significant strengthening of previous "lim sup divergence" results. For $H^1$, they show that for any sequence of sampling points, there exists a function in $H^1$ for which the reconstruction diverges in the lim sup sense. The proofs often rely on constructing specific sequences of functions and then analyzing their behavior using properties of entire functions, Fourier analysis, and inequalities related to sampling series. They also leverage concepts like Riesz bases and interpolating sequences.
Datasets
Not discussed in the paper.
Key Insights and Ideas
- Strong Divergence is Inherent: The most significant insight is that divergence in these reconstruction methods isn't just a rare occurrence for specific subsequences; it's a fundamental property that can manifest as strong divergence in the limit for certain signals. This rules out the possibility of simply choosing "good" subsequences to guarantee convergence.
- PW vs. H¹ Behavior: While strong divergence is shown for $PW_\pi^1$, the results for $H^1$ are slightly different, showing lim sup divergence for any subsequence of sampling points, but with the possibility of finding a subsequence for which a specific function converges (a question posed as Conjecture 3.2).
- Oscillatory Behavior: The paper highlights that when divergence occurs, the reconstruction series can exhibit strong oscillations, with the maximum and minimum values tending towards positive and negative infinity, respectively. This provides a qualitative understanding of the nature of the divergence.
- Beyond Banach-Steinhaus: The authors emphasize that their results go beyond what can be achieved with the classical Banach-Steinhaus Theorem, which typically deals with uniform boundedness of linear operators and doesn't directly address strong divergence of specific function reconstructions.
Key Results
- Theorem 2.1: There exists a function in $PW_\pi^1$ for which the Shannon sampling series diverges strongly (error tends to $+\infty$).
- Theorem 3.1: For any sequence of natural numbers, there exists a function in $H^1$ for which a specific reconstruction procedure exhibits lim sup divergence.
- Corollary 2.3: For any fixed point $t_0$, there exists a function in $PW_\pi^1$ such that a Valiron interpolation series diverges strongly.
- Theorem 4.2: For functions of sine type in $PW_\pi^1$, there exists a function whose reconstruction using its zeros as sampling points diverges strongly.
- Theorem 5.1: A specific reconstruction method for generating functions in $PW_\pi^1$ exhibits strong divergence.
- Theorem 6.1: The reconstruction sequences from Theorems 4.2 and 5.1 not only diverge in maximum absolute value but also exhibit strong divergence of both their maximum and minimum values.
Limitations and Future Work
- Universality of Divergence: While the paper shows divergence for some functions, a key open question (Conjecture 2.5) is whether there exists a universal sequence of sampling points for which all functions in $PW_\pi^1$ would have bounded reconstructions. The paper suggests this is unlikely.
- H¹ Conjecture: Conjecture 3.2 posits that for any function in $H^1$, there exists a subsequence of sampling points that leads to bounded reconstruction. This suggests that adaptive sampling strategies might be effective for $H^1$.
- Characterizing Functions for Divergence: Question 6.3 asks for a characterization of functions $\psi(N)$ for which the Shannon series reconstruction can diverge strongly for some function in $PW_\pi^1$. This points towards further research into the specific properties of signals that lead to reconstruction failure.
- Quantitative Estimates: The paper mentions that while rates of divergence are not explicitly stated, the proofs are constructive and could potentially yield quantitative estimates.
Summary updated