Skip to main content

Probability inequality concerning a chirp-type signal

Abstract

Probability inequalities of random variables play important roles, especially in the theory of limiting theorems of sums of independent random variables. In this note, we give a proof of the following:

Let { ϵ t } be a sequence of independent random variables such that E( ϵ t )=0, E( ϵ t 2 )=v<∞ and E( ϵ t 4 )=σ<∞. Then

max 0 ≤ ω < π 0 ≤ α < π | ∑ t = 1 n t β ϵ t e i ( ω t + α t 2 ) | 2 = O p ( n β + 7 8 ) .

It is shown that this result will be useful in estimating the parameters of a chirp-type statistical model and in establishing the consistency of estimators.

1 Introduction

In [1] Whittle (1952) considered the problem of estimating the parameters of a sine wave:

X t =Acosωt+Bsinωt+ ϵ t ,

where X t ’s are the observations and ϵ t ’s are independent, identically distributed random variables with mean zero and finite(unknown) variance. Whittle’s solution to the problem of estimating the parameters and the proof of consistency of parameters of the above sine wave used arguments which are not mathematically rigorous. In 1973, a rigorous solution to Whittle’s problem was given by Walker [2]. In his proof of consistency of parameters, he used the following O p result:

max 0 ≤ ω ≤ π | ∑ t = 1 n Y t e i ω t |= O p ( n 3 4 ) ,

where { Y t } is a linear process.

In this paper, we extend the above O p result to obtain

max 0 ≤ ω < π 0 ≤ α < π | ∑ t = 1 n t β ϵ t e i ( ω t + α t 2 ) | 2 = O p ( n β + 7 8 ) .
(1)

We came across this problem while attempting to establish the consistency of the parameters of the model

X t =Acos ( ω t + α t 2 ) +Bsin ( ω t + α t 2 ) + ϵ t ,

where X t ’s and ϵ t ’s are as mentioned above. Models of this type are referred to as ‘chirp’ models [3, 4] and [5], and they have drawn the attention of many researchers [6, 7] and [8]. Although we tried to find our main result given in (1) in the literature, we were unable to find one. We make use of the following basic results to establish our main result.

2 Basic results

Definition 1 Let { X n } be a sequence of random variables. We say that X n = O p (1) if for given ϵ>0, there exists M such that

P [ | X n | > M ] <ϵfor all n.

Let ( a n ) be a sequence of nonzero real numbers. We say that X n = O p ( a n ) if X n / a n = O p (1).

Lemma 1 Let { X n } be a sequence of random variables, and let { a n } and { b n } be two sequence of nonzero real numbers. Then

X n = O p ( b n )⇒ a n X n = O p ( a n b n ).

Lemma 2 Let { X n } be a sequence of random variables, let { a n } be a sequence of real numbers, and let q be a positive real number. Then

X n = O p ( a n )⇒ X n q = O p ( a n q ) .

Definition 2 Let { X n } be a sequence of random variables. We say that X n = o p (1) if X n →0 in probability. (So, X n = o p (1) if for each ϵ≥0, we have lim n → ∞ P(| X n |>ϵ)=0.)

Lemma 3 Let { X n } be a sequence of random variables, and let { a n } be a sequence of real numbers with a n →0 as n→∞. If X n = O p (1), then a n X n ⟶ P 0.

Lemma 4 Let { X n } be a sequence of random variables, and let β be a positive real number. Then

X n = O p ( 1 n β ) ⇒ X n = o p (1).

Lemma 5 Let { X n } be a sequence of random variables, and let β be any real number. If X n = O p ( n β ) and Y n = O p ( n β ), then X n + Y n = O p ( n β ).

Lemma 6 Let μ be any real number. Let { X n } be a sequence of random variables such that E(| X n |)=O( n μ ). Then X n =O( n μ ).

We now establish two lemmas that we need to prove the main result.

Lemma 7 Let a t , u ; t=1,2,…,n, u=1,2,…,n be complex numbers. Then

∑ t = 1 n ∑ u = 1 n a t , u = ∑ s = 1 n − 1 ∑ t = 1 n − s [ a t , t + s + a t + s , t ]+ ∑ t = 1 n a t , t .

Proof The ∑ t = 1 n ∑ u = 1 n a t , u is the sum of the terms in the table:

a 1 , 1 a 1 , 2 a 1 , 3 ⋯ a 1 , n − 1 a 1 , n a 2 , 1 a 2 , 2 a 2 , 3 ⋯ a 2 , n − 1 a 2 , n ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ a n − 1 , 1 a n − 1 , 2 a n − 1 , 3 ⋯ a n − 1 , n − 1 a n − 1 , n a n , 1 a n , 2 a n , 3 ⋯ a n , n − 1 a n , n .

We can add by summing along the diagonals. So, the sum is equal to

∑ s = 1 n − 1 ∑ t = 1 n − s a t , t + s + ∑ p = 1 n − 1 ∑ u = 1 n − p a u + p , u + ∑ t = 1 n a t , t .

Therefore

∑ t = 1 n ∑ u = 1 n a t , u = ∑ s = 1 n − 1 ∑ t = 1 n − s [ a t , t + s + a t + s , t ]+ ∑ t = 1 n a t , t .

 □

Lemma 8 Let b t ; t=1,2,…,n be complex numbers. Then

| ∑ t = 1 n b t | 2 ≤2| ∑ s = 1 n − 1 ∑ t = 1 n − s b t b ¯ t + s |+ ∑ t = 1 n | b t | 2 .

Proof Since | ∑ t = 1 n b t | 2 = ∑ t = 1 n ∑ u = 1 n b t b ¯ u , by using a t , u = b t b ¯ u in Lemma 7, then

| ∑ t = 1 n b t | 2 = ∑ s = 1 n − 1 ∑ t = 1 n − s [ b t b ¯ t + s + b t + s b ¯ t ] + ∑ t = 1 n b t b ¯ t = ∑ s = 1 n − 1 ∑ t = 1 n − s 2 ℜ b t b ¯ t + s + ∑ t = 1 n | b t | 2 ≤ 2 | ∑ s = 1 n − 1 ∑ t = 1 n − s b t b ¯ t + s | + ∑ t = 1 n | b t | 2 .

 □

3 Main result

Theorem 1 Let { ϵ t } be a sequence of independent random variables such that E( ϵ t )=0, E( ϵ t 2 )=v<∞ and E( ϵ t 4 )=σ<∞. Then

max 0 ≤ ω < π 0 ≤ α < π | ∑ t = 1 n t β ϵ t e i ( ω t + α t 2 ) | 2 = O p ( n β + 7 8 ) .

Proof Letting b t = t β ϵ t e i ( ω t + α t 2 ) in Lemma 8, we see that

| ∑ t = 1 n t β ϵ t e i ( ω t + α t 2 ) | 2 ≤ 2 | ∑ s = 1 n − 1 ∑ t = 1 n − s t β ( t + s ) β ϵ t ϵ t + s e i ( ω t + α t 2 ) e − i ( ω ( t + s ) + α ( t + s ) 2 ) | + ∑ t = 1 n | t β ϵ t e i ( ω t + α t 2 ) | 2 = 2 | ∑ s = 1 n − 1 ∑ t = 1 n − s t β ( t + s ) β ϵ t ϵ t + s e − i ω s e − i α s 2 e − 2 i α t s | + ∑ t = 1 n t 2 β ϵ t 2 .

Using the triangle inequality, we obtain

| ∑ t = 1 n t β ϵ t e i ( ω t + α t 2 ) | 2 ≤2 ∑ s = 1 n − 1 | ∑ t = 1 n − s t β ( t + s ) β ϵ t ϵ t + s e − 2 i α t s |+ ∑ t = 1 n t 2 β ϵ t 2 .
(2)

Fix s and consider | ∑ t = 1 n − s t β ( t + s ) β ϵ t ϵ t + s e − 2 i α t s |. By Lemma 8 with b t = t β ( t + s ) β ϵ t ϵ t + s e − 2 i α t s , we obtain

| ∑ t = 1 n − s t β ( t + s ) β ϵ t ϵ t + s e − 2 i α t s | 2 ≤ 2 | ∑ p = 1 n − s − 1 ∑ t = 1 n − s − p t β ( t + s ) β ( t + p ) β ( t + p + s ) β ϵ t ϵ t + s e − 2 i α t s ϵ t + p ϵ t + p + s e 2 i α s ( t + p ) | + ∑ t = 1 n − s | t β ( t + s ) β ϵ t ϵ t + s e − 2 i α t s | 2 ,

and it follows from the triangle inequality that

| ∑ t = 1 n − s t β ( t + s ) β ϵ t ϵ t + s e − 2 i α t s | 2 ≤ 2 ∑ p = 1 n − s − 1 | ∑ t = 1 n − s − p t β ( t + s ) β ( t + p ) β ( t + p + s ) β ϵ t ϵ t + s ϵ t + p ϵ t + p + s | + ∑ t = 1 n − s t β ( t + s ) β ϵ t 2 ϵ t + s 2 .

Therefore

| ∑ t = 1 n − s t β ( t + s ) β ϵ t ϵ t + s e − 2 i α t s | 2 ≤ 2 [ ∑ p = 1 n − s − 1 | ∑ t = 1 n − s − p t β ( t + s ) β ( t + p ) β ( t + p + s ) β ϵ t ϵ t + s ϵ t + p ϵ t + p + s | + ∑ t = 1 n − s t 2 β ( t + s ) 2 β ϵ t 2 ϵ t + s 2 ] 1 / 2 .

Substituting this in equation (2), we obtain

| ∑ t = 1 n t β ϵ t e i ( ω t + α t 2 ) | 2 ≤ 2 ∑ s = 1 n − 1 [ 2 ∑ p = 1 n − s − 1 | ∑ t = 1 n − s − p t β ( t + s ) β ( t + p ) β ( t + p + s ) β ϵ t ϵ t + s ϵ t + p ϵ t + p + s | + ∑ t = 1 n − s t 2 β ( t + s ) 2 β ϵ t 2 ϵ t + s 2 ] 1 / 2 + ∑ t = 1 n t 2 β ϵ t 2 .

It follows that

max 0 ≤ ω < π 0 ≤ α < π | ∑ t = 1 n t β ϵ t e i ( ω t + α t 2 ) | 2 ≤ 2 2 ∑ s = 1 n − 1 [ ∑ p = 1 n − s − 1 | ∑ t = 1 n − s − p t β ( t + s ) β ( t + p ) β ( t + p + s ) β ϵ t ϵ t + s ϵ t + p ϵ t + p + s | + ∑ t = 1 n − s t 2 β ( t + s ) 2 β ϵ t 2 ϵ t + s 2 ] 1 / 2 + ∑ t = 1 n t 2 β ϵ t 2 .

Taking the expectation of both sides of the above inequality, we obtain

E [ max 0 ≤ ω < π 0 ≤ α < π | ∑ t = 1 n t β ϵ t e i ( ω t + α t 2 ) | 2 ] ≤ 2 2 ∑ s = 1 n − 1 E [ ∑ p = 1 n − s − 1 | ∑ t = 1 n − s − p t β ( t + s ) β ( t + p ) β ( t + p + s ) β ϵ t ϵ t + s ϵ t + p ϵ t + p + s | + ∑ t = 1 n − s t 2 β ( t + s ) 2 β ϵ t 2 ϵ t + s 2 ] 1 / 2 + ∑ t = 1 n t 2 β E ( ϵ t 2 ) .

We claim that

E [ max 0 ≤ ω < π 0 ≤ α < π | ∑ t = 1 n t β ϵ t e i ( ω t + α t 2 ) | 2 ] =O ( n 2 β + 7 / 4 ) .

We note that since

∑ t = 1 n t 2 β E ( ϵ t 2 ) = ∑ t = 1 n t 2 β v=O ( n 2 β + 1 ) ≤O ( n 2 β + 7 / 4 ) ,

we can ignore the term ∑ t = 1 n t 2 β E( ϵ t 2 ) provided, and we show the other term is O( n 2 β + 7 / 4 ).

Taking x= | z | 1 / 2 in Schwarz’s inequality E|x|≤ ( E | x | 2 ) 1 / 2 yields E( | z | 1 / 2 )≤ ( E | z | ) 1 / 2 . Using this on the right-hand side of the above inequality, we get

E [ max 0 ≤ ω < π 0 ≤ α < π | ∑ t = 1 n t β ϵ t e i ( ω t + α t 2 ) | 2 ] ≤ ∑ s = 1 n − 1 { ∑ p = 1 n − s − 1 E | ∑ t = 1 n − s − p t β ( t + s ) β ( t + p ) β ( t + p + s ) β ϵ t ϵ t + s ϵ t + p ϵ t + p + s | + ∑ t = 1 n − s t 2 β ( t + s ) 2 β E ( ϵ t 2 ϵ t + s 2 ) } 1 / 2 .

Using Schwarz’s inequality again on the first term on the right, we get

E [ max 0 ≤ ω < π 0 ≤ α < π | ∑ t = 1 n t β ϵ t e i ( ω t + α t 2 ) | 2 ] ≤ ∑ s = 1 n − 1 [ ∑ p = 1 n − s − 1 [ E { ∑ t = 1 n − s − p t β ( t + s ) β ( t + p ) β ( t + p + s ) β ϵ t ϵ t + s ϵ t + p ϵ t + p + s 2 } 2 ] 1 / 2 + ∑ t = 1 n − s t 2 β ( t + s ) 2 β E ( ϵ t 2 ϵ t + s 2 ) ] 1 / 2 = ∑ s = 1 n − 1 [ ∑ p = 1 n − s − 1 [ E ( ∑ t = 1 n − s − p t 2 β ( t + s ) β ( t + p ) 2 β ( t + p + s ) 2 β ϵ t 2 ϵ t + s 2 ϵ t + p 2 ϵ t + p + s 2 ) + E ( 2 ∑ u , v = 1 u < v n − s − p u β ( u + s ) β ( u + p ) β ( u + p + s ) β v β ( v + s ) β ( v + p ) β ( v + p + s ) β ϵ u ϵ u + s ϵ u + p ϵ u + p + s ϵ v ϵ v + s ϵ v + p ϵ v + p + s ) ] 1 / 2 + ∑ t = 1 n − s t 2 β ( t + s ) 2 β E ( ϵ t 2 ϵ t + s 2 ) ] 1 / 2 .

So,

E [ max 0 ≤ ω < π 0 ≤ α < π | ∑ t = 1 n t β ϵ t e i ( ω t + α t 2 ) | 2 ] ≤ ∑ s = 1 n − 1 [ ∑ p = 1 n − s − 1 [ ∑ t = 1 n − s − p t 2 β ( t + s ) 2 β ( t + p ) 2 β ( t + p + s ) 2 β E ( ϵ t 2 ϵ t + s 2 ϵ t + p 2 ϵ t + p + s 2 ) + 2 ( ∑ u , v = 1 u < v n − s − p u β ( u + s ) β ( u + p ) β ( u + p + s ) β v β ( v + s ) β ( v + p ) β ( v + p + s ) β E ( ϵ u ϵ u + s ϵ u + p ϵ u + p + s ϵ v ϵ v + s ϵ v + p ϵ v + p + s ) ) ] 1 / 2 + ∑ t = 1 n − s t 2 β ( t + s ) 2 β E ( ϵ t 2 ϵ t + s 2 ) ] 1 / 2 .

Consider E( ϵ u ϵ u + s ϵ u + p ϵ u + p + s ϵ v ϵ v + s ϵ v + p ϵ v + p + s ). Since u≤v, v+p+s is the unique largest subscript, the ϵ t ’s are independent and E( ϵ t )=0, we have

E( ϵ u ϵ u + s ϵ u + p ϵ u + p + s ϵ v ϵ v + s ϵ v + p ϵ v + p + s )=E( ϵ u ϵ u + s ϵ u + p ϵ u + p + s ϵ v ϵ v + s ϵ v + p )E( ϵ v + p + s )=0.

Therefore

E [ max 0 ≤ ω < π 0 ≤ α < π | ∑ t = 1 n t β ϵ t e i ( ω t + α t 2 ) | 2 ] ≤ ∑ s = 1 n − 1 [ ∑ p = 1 n − s − 1 [ ∑ t = 1 n − s − p t 2 β ( t + s ) 2 β ( t + p ) 2 β ( t + p + s ) 2 β E ( ϵ t 2 ϵ t + s 2 ϵ t + p 2 ϵ t + p + s 2 ) ] 1 / 2 + ∑ t = 1 n − s t 2 β ( t + s ) 2 β E ( ϵ t 2 ϵ t + s 2 ) ] 1 / 2 .

Now consider the term E( ϵ t 2 ϵ t + s 2 ). Since E( ϵ t 2 )=v≤∞, and the ϵ t ’s are independent,

∑ t = 1 n − s t 2 β ( t + s ) 2 β E ( ϵ t 2 ϵ t + s 2 ) ≤K n 4 β + 1 .

Similarly, since E( ϵ t 2 )=v≤∞, E( ϵ t 4 )=σ≤∞, and the ϵ t ’s are independent,

∑ t = 1 n − s − p t 2 β ( t + s ) 2 β ( t + p ) 2 β ( t + p + s ) 2 β E ( ϵ t 2 ϵ t + s 2 ϵ t + p 2 ϵ t + p + s 2 ) =K n 8 β + 1 .

Thus

E [ max 0 ≤ ω < π 0 ≤ α < π | ∑ t = 1 n t β ϵ t e i ( ω t + α t 2 ) | 2 ] ≤ ∑ s = 1 n − 1 [ ∑ p = 1 n − s − 1 [ K n 8 β + 1 ] 1 / 2 + K n 4 β + 1 ] 1 / 2 = ∑ s = 1 n − 1 [ n K n 4 β + 1 / 2 + K n 4 β + 1 ] 1 / 2 = ∑ s = 1 n − 1 [ K n 4 β + 3 / 2 ] 1 / 2 = n K n 2 β + 3 / 4 = O ( n 2 β + 7 / 4 ) .

It follows by virtue of Lemma 6 that

max 0 ≤ ω < π 0 ≤ α < π | ∑ t = 1 n t β ϵ t e i ( ω t + α t 2 ) | 2 = O p ( n 2 β + 7 / 4 ) .

Using Lemma 3, we obtain

max 0 ≤ ω < π 0 ≤ α < π | ∑ t = 1 n t β ϵ t e i ( ω t + α t 2 ) |= O p ( n β + 7 / 8 ) ,

which completes the proof of the theorem. □

References

  1. Whittle P: The simultaneous estimation of a time series harmonic component and covariance structure. Trab. Estad. 1952, 3: 43–57. 10.1007/BF03002861

    Article  MathSciNet  Google Scholar 

  2. Walker AM: On the estimation of a harmonic component in a time series with stationary dependent residuals. Adv. Appl. Probab. 1973, 5: 217–241. 10.2307/1426034

    Article  Google Scholar 

  3. Bretthorst GL: Lecture Notes in Statistics. Springer, Berlin; 1965.

    Google Scholar 

  4. Smith CR, Erickson GJ: Maximum-Entrophy and Bayesian Spectral Analysis and Estimation Problems. Reidel, Dordrecht; 1987.

    Book  Google Scholar 

  5. Kundu D, Nandi S: Parameter estimation of chirp signals in presence of stationary noise. Stat. Sin. 2008, 18: 187–201.

    MathSciNet  Google Scholar 

  6. Cochrane T, Pinner C: An improved Mordell type bound for exponential sums. Proc. Am. Math. Soc. 2004, 133(2):313–320.

    Article  MathSciNet  Google Scholar 

  7. Konyagin, SV, Lev, VF: On the Distribution of Exponential Sums. Electron. J. Comb. Number Theory 0, A01 (2000)

  8. Bourgain J, Glibichuk AA, Konyagin SV: Estimates for the number of sums and products and for exponential sums in fields of prime order. J. Lond. Math. Soc. 2005, 73(2):380–398.

    Article  MathSciNet  Google Scholar 

Download references

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Kanthi Perera.

Additional information

Competing interests

The author declares that she has no competing interests.

Rights and permissions

Open Access This article is distributed under the terms of the Creative Commons Attribution 2.0 International License (https://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Reprints and permissions

About this article

Cite this article

Perera, K. Probability inequality concerning a chirp-type signal. J Inequal Appl 2013, 131 (2013). https://doi.org/10.1186/1029-242X-2013-131

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/1029-242X-2013-131

Keywords