ProbXiv
Explorersign in
Problem archiveProblem record

Statement

Let a1,a2,a3,…a_1, a_2, a_3, \ldots be a sequence of real numbers (a signal) transmitted in the presence of additive noise such that a1+ϵ1,a2+ϵ2,a3+ϵ3,…a_1 + \epsilon_1, a_2 + \epsilon_2, a_3 + \epsilon_3, \ldots is received (both are infinite sequences). We observe X1,X2,X3,…X_1, X_2, X_3, \ldots which is \emph{either} pure noise ϵ1,ϵ2,ϵ3,…\epsilon_1, \epsilon_2, \epsilon_3, \ldots or the transmitted signal a1+ϵ1,a2+ϵ2,a3+ϵ3,…a_1 + \epsilon_1, a_2 + \epsilon_2, a_3 + \epsilon_3, \ldots. To decide which of the following is received:

(ϵ1,ϵ2,ϵ3,…)∼p(R∞),(a1+ϵ1,a2+ϵ2,a3+ϵ3,…)∼p(R∞),\begin{split} (\epsilon_1, \epsilon_2, \epsilon_3, \ldots) &\sim p(\mathbb{R}^{\infty}),\\ (a_1 + \epsilon_1, a_2 + \epsilon_2, a_3 + \epsilon_3, \ldots) &\sim p(\mathbb{R}^{\infty}), \end{split}

we would like to know whether the measures generated by the sequences above are mutually singular. It is known that if the signal is “strong enough", meaning

∑i=1∞ai2=∞,\sum_{i=1}^\infty a_i^2 = \infty,

then detection is possible. Larry Shepp also noticed that if ϵi\epsilon_i are i.i.d.\ having a pdf such that I(ϵi)I(\epsilon_i) is finite, then the above “strong enough" property is also necessary.

The question is the following. Suppose the signal is weak, i.e.

∑i=1∞ai2<∞,\sum_{i=1}^\infty a_i^2 < \infty,

but

∑i=1∞∣ai∣λ=∞,\sum_{i=1}^\infty |a_i|^{\lambda} = \infty,

for some λ∈(0,2)\lambda \in (0,2). If ϵi∼F\epsilon_i \sim F, what is the infimum of such λ\lambda for which every signal with ∑i=1∞∣ai∣λ=∞\sum_{i=1}^\infty |a_i|^{\lambda} = \infty is detectable.

In the case [−1,+1][-1, +1], the problem has been studied and it is known that the smallest λ\lambda such that detection is possible is λ=1\lambda = 1.

If ϵi∼U(−a,+a)\epsilon_i \sim U(-a, +a), it seems that if P(ϵi=±1)=12P(\epsilon_i = \pm 1) = \frac{1}{2}, a signal with ∑i=1∞∣ai∣λ=∞\sum_{i=1}^\infty |a_i|^{\lambda} = \infty for some λ>0\lambda > 0 is detectable so that in this case the infimum equals 00.

Record

Comments

Nothing has been published against this problem yet, and nothing has been said about it.

Sign in with an institutional address to take part in the discussion. Reading every thread stays open to everyone.

Sign in

Solve with an agent

Open the statement in a chat, with the problem and the ground rules already written into the prompt.

This opens a third-party site. Nothing is posted back to ProbXiv and nothing you write there is recorded here — what a model gives you is an attempt, which a person still has to check.