Bounds on Asymptotic Merit Factor of Chu Sequences

Idris Mercer, University of Delaware, idmercer@math.udel.edu

Abstract

Chu sequences are a family of polyphase sequences that have perfect periodic autocorrelations and good aperiodic autocorrelations. It has previously been proved that the maximum opeak (aperiodic) autocorrelation (in absolute value) of the Chu sequence of length n is asymptotically equal to 0.480261 n. It has also been empirically observed that the merit factor of Chu sequences appears to grow like a constant times n. In this note, we provide an analytic proof that the merit factor the Chu sequence of length n is bounded below of by a constant multiple of n for all n.

Preliminaries

A polyphase sequence of length n is a nite sequence S = (s0 , s1 , . . . , sn1 ), where for each j, we have sj C and |sj | = 1. For k = 0, 1, . . . , n 1, we dene the aperiodic autocorrelations of S,
nk1

ck =
j=0

sj sj+k ,

and the periodic autocorrelations of S,

n1

k =
j=0

sj sj+k ,

where the bar denotes complex conjugation. In the denition of k , the addition in the subscript is modulo n. 1

We have c0 = 0 = n, which we call the trivial autocorrelations. It is not hard to show that k = ck + cnk . Also note that |cn1 | = |s0 sn1 | = 1. We can either seek polyphase sequences with periodic autocorrelations near zero, or with aperiodic autocorrelations near zero. If k = 0 for all k = 0, we say S is a perfect sequence. If |ck | 1 for all k = 0, we say S is a polyphase Barker sequence. Given a polyphase sequence S, two natural measures of the closeness to zero of the aperiodic autocorrelations are as follows: PSL(S) = max |ck | ,
k=0

TSE(S) =
k=0

|ck |2 ,

which we respectively call the peak sidelobe level and total sidelobe energy. (Some authors dene the TSE to be twice our value, since they also dene ck = ck .) One can also dene the merit factor of the sequence S, which in our notation is MF(S) = We dene three functions of n: PSLmin (n) = min PSL(S),
S S

n2 . 2 TSE(S)

TSEmin (n) = min TSE(S), MFmax (n) = max MF(S) =

S

n2 , 2 TSEmin (S)

where the extremum is taken over all polyphase sequences of length n. These are mathematically well-dened (the space of all polyphase sequences of length n is the product of n copies of the unit circle, and is hence compact) but explicit computation of the extrema appears very dicult. It would be of interest to have good bounds for the growth rates of the functions PSLmin (n) and TSEmin (n). 2

If S is a polyphase Barker sequence of length n, then PSL(S) = 1 and TSE(S) n 1. Therefore, if there exists an innite family of polyphase Barker sequences, their merit factor would be at least n2 /(2(n 1)) n/2. Polyphase Barker sequences have been found for all lengths up to N , where the value of N has been gradually increasing. For example, Friese [4] found polyphase Barker sequences of all lengths up to 36, but conjectured that they do not exist for signicantly higher lengths. However, Borwein and Ferguson [2] found polyphase Barker sequences of all lengths up to 63. It is unknown whether there exist polyphase Barker sequences for every length n. Thus, we have PSLmin (n) = 1 for all n 63, and possibly for larger n. However, the best known asymptotic upper bound in the literature appears to be of the form PSLmin (n) = O( n). See, for example, Mow and Li [5]. It has been empirically observed [1, 6] that Chu sequences, and some related sequences, appear to have merit factor that grows like a constant times n, or equivalently, their TSE appears to grow like a constant times n3/2 . The main purpose of this note is to provide an analytic proof that, if S is the Chu sequence of length n, we have a bound of the form TSE(S) =
k=0

|ck |2 Bn3/2

that holds for all n (where B is constant). This implies TSEmin (n) Bn3/2 .

Proof of Main Result

We consider the sequences dened in [3]. For each positive integer n, the Chu sequence of length n is the sequence S = (s0 , s1 , . . . , sn1 ), where 2 j e if n is even, 2n sj = j(j +1) e if n is odd, 2n 3

where e(t) is shorthand for ei2t . Assume k = 0. As proved in [3], the Chu sequences satisfy k = 0 for all k. This implies that |cnk | = |ck |. Now note that in general, cnk is a sum of k terms of absolute value at most 1. Therefore |ck | = |cnk | k. Using some relatively straightforward manipulations (see, e.g., the beginning of Section IV in [5], or the proof of Theorem 2 in [7]) one can show that the Chu sequences satisfy k 2 sin n . |ck | = k sin n For convenience, we dene the half-energy HE(S) =
0<k< n/2

|ck |2 .

Using the relation |cnk | = |ck |, one can see that TSE(S) = Now note that sin cn/2 = n 4 sin 2 = sin n 4 1. 2 HE(S) 2 HE(S) + cn/2
2

if n is odd, if n is even.

The key is now to break the sum HE(S) into two pieces HE(S) =
1k An1/2

|ck |2 +
An1/2 <k< n/2

|ck |2

where A is a xed positive real number yet to be determined. Let m = An1/2 and let m2 = n/2 1. Since |ck | k, we have |ck |2
1km 1km

k2 =

m3 m2 m + + . 3 2 6

We now consider |ck |2 =

m<k< n/2 m<km2

sin2

k 2 k csc2 n n

m<km2

csc2

k . n

This latter sum is equal to n csc2

m<km2

k 1 n n

which can be regarded as a Riemann sum. If the interval of real numbers [ m , m2 ] is divided into subintervals of width 1/n, then as k ranges through n n the integer values m < k m2 , the numbers k/n are the right endpoints of the subintervals. Note that m2 (n 1)/2 < n/2, and that m > 0. Therefore [ m , m2 ] n n 1 is a subset of the interval (0, 2 ), on which the function csc2 (t) is strictly decreasing. It follows that we have n csc2
m<km2

1 k n n

m2 /n

n csc2 (t) dt
m/n

and the integral evaluates to n cot(t)

m2 /n

=
m/n

n m m2 cot cot n n

(1)

On the interval (0, ), the function cot x is positive and satises cot x < 1/x. 2 The quantity (1) can therefore be bounded above by n m n n n2 cot < = 2 . n m m Combining our bounds, we conclude HE(S) < n2 m3 m2 m + + + 2 . 3 2 6 m

Since A n m < A n + 1, we conclude (A n + 1)3 (A n + 1)2 A n + 1 n2 HE(S) < + + + 2 3 2 6 A n 5

(2)

which is a polynomial in power n3/2 is

n of degree 3. The coecient of the dominant

1 A3 + 2 . 3 A Using elementary calculus, we nd that the positive A that minimizes this quantity is 1 A= and the minimum value itself is A3 1 4 + 2 = 3/2 0.2394494961. 3 A 3 Choosing A = 1/ , we then have an upper bound of the form HE(S) < which implies TSE(S) 2 HE(S) + 1 < 8 n3/2 + O(n) 3 3/2 4 n3/2 + O(n) 3 3/2

0.4788989922 n3/2 + O(n). This implies that the merit factor MF(S) asymptotically grows at least like n2 3 3/2 = n 1.044061500 n. 8 16 2 3/2 n3/2 3 If we seek explicit bounds that are valid for all n, we can start by evaluating the bound (2) when A = 1/ . This gives HE(S) < 4 3 13 n3/2 + n + n1/2 + 1 3/2 3 2 6

and then since TSE(S) 2 HE(S) + 1, we have TSE(S) < 8 3 13 n3/2 + n + n1/2 + 3. 3/2 3 3 6

A computation then reveals that we have, for example, TSE(S) < 3.437207165 n3/2 for all n 2, TSE(S) < 0.8482098513 n3/2 for all n 20, TSE(S) < 0.5597074838 n3/2 for all n 200, TSE(S) < 0.5015078204 n3/2 for all n 2000, TSE(S) < 0.4857746665 n3/2 for all n 20000, which implies MF(S) > 0.1454669376 n for all n 2, MF(S) > 0.5894767660 n for all n 20, MF(S) > 0.8933237710 n for all n 200, MF(S) > 0.9969934260 n for all n 2000, MF(S) > 1.029283811 n for all n 20000. Antweiler and Bmer [1] observed that the merit factor of Chu sequences o appears empirically to grow like a constant times n, but to the best of the current authors knowledge, the existing literature does not contain proofs that the Chu sequences satisfy bounds of the form TSE(S) < Bn3/2 and MF(S) > C n that hold for all n.

Conclusions and Further Questions

We have proved that the Chu sequences satisfy, in an asymptotic sense, TSE(S) MF(S) 0.4788989922 n3/2 , 1.044061500 n.

Numerical evidence suggests the constants in front are not the best constants. If we recall that
n1 n1

TSE(S) =
k=1

|ck | =
k=1

sin2 7

k k 2 csc2 , n n

then we can calculate TSE(n) numerically for certain values of n and compare with n3/2 . Doing this, for example, with n from 1000 to 1024 suggests TSE(S) 0.318 n3/2 , MF(S) 1.57 n. The empirical result MF(S) 1.57 n was already noted in [1]. Furthermore, it was shown in [6] that certain other sequences, related but not identical to Chu sequences, appear to satisfy MF(S) C n for a larger constant C. We have shown that the function TSEmin (n) is bounded above by a constant times n3/2 . It would be interesting to know whether this can be improved. If there are polyphase Barker sequences of every length, then TSEmin (n) grows at most linearly in n.

References
[1] M. Antweiler & L. Bmer, Merit factor of Chu and Frank sequences, o Electron. Lett. 26 (1990), 20682070. [2] P. Borwein & R. Ferguson, Polyphase sequences with low autocorrelation, IEEE Trans. Inform. Theory 51 (2005), 15641567. [3] D. C. Chu, Polyphase codes with good periodic correlation properties, IEEE Trans. Inform. Theory 14 (1972), 531532. [4] M. Friese, Polyphase Barker sequences up to length 36, IEEE Trans. Inform. Theory 42 (1996), 12481250. [5] W. H. Mow & S.-Y. R. Li, Aperiodic autocorrelation and crosscorrelation of polyphase sequences, IEEE Trans. Inform. Theory 43 (1997), 1000 1007. [6] P. Rapajic & R. Kennedy, Merit factor based comparison of new polyphase sequences, IEEE Commun. Lett. 2 (1998), 269270. [7] N. Zhang & S. Golomb, Polyphase sequence with low autocorrelations, IEEE Trans. Inform. Theory 39 (1993), 10851089.