1

Five infinite families of binary cyclic codes and

their related codes with good parameters
Hai Liu, Chengju Li, Cunsheng Ding
arXiv:2301.06446v1 [cs.IT] 16 Jan 2023

Abstract

Cyclic codes are an interesting type of linear codes and have wide applications in communication
and storage systems due to their efficient encoding and decoding algorithms. Inspired by the recent
work on binary cyclic codes published in IEEE Trans. Inf. Theory, vol. 68, no. 12, pp. 7842-7849,
2022, the objectives of this paper are the construction and analyses of five infinite families of binary
cyclic codes with parameters [n, k] and (n− 6)/3 ≤ k ≤ 2(n+ 6)/3. Three of the five families of binary
cyclic codes and their duals have a very good lower bound on their minimum distances and contain
distance-optimal codes. The other two families of binary cyclic codes are composed of binary duadic
codes with a square-root-like lower bound on their minimum distances. As a by-product, two infinite
families of self-dual binary codes with a square-root-like lower bound on their minimum distances are
obtained.

Keywords Cyclic code; duadic code; linear code, self-dual code.

The work of Hai Liu and Chengju Li was supported by the National Natural Science Foundation of China (12071138),
Shanghai Rising-Star Program (22QA1403200), the open research fund of National Mobile Communications Research Laboratory
of Southeast University (2022D05), and the Shanghai Trusted Industry Internet Software Collaborative Innovation Center. The
work of Cunsheng Ding was supported by the Hong Kong Research Grants Council under Grant 16302121. (Corresponding
author: Chengju Li.)
H. Liu and C. Li are with the Shanghai Key Laboratory of Trustworthy Computing, East China Normal University, Shanghai,
200062, China; and are also with the National Mobile Communications Research Laboratory, Southeast University, Nanjing
210096, China (email: [email protected], [email protected]).
C. Ding is with the Department of Computer Science and Engineering, The Hong Kong University of Science and Technology,
Clear Water Bay, Kowloon, Hong Kong, China (email: [email protected]).
 2

C ONTENTS

1 Introduction 3

2 The first three families of binary cyclic codes and their duals 5
2-A The construction of the first three families of binary cyclic codes . . . . . 5
2-B Some auxiliary results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2-B1 The odd m case . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2-B2 The even m case . . . . . . . . . . . . . . . . . . . . . . . . . 9
2-C Parameters of the binary codes C(i,m) . . . . . . . . . . . . . . . . . . . . . 11
2-C1 Parameters of C(i,m) when m ≥ 5 is odd . . . . . . . . . . . . . 11
2-C2 Parameters of C(i,m) when m ≥ 4 is even . . . . . . . . . . . . 13
⊥
2-D Parameters of the dual codes C(i,m) . . . . . . . . . . . . . . . . . . . . . . 15
⊥
2-D1 Parameters of C(i,m) when m ≥ 5 is odd . . . . . . . . . . . . . 16
⊥
2-D2 Parameters of C(i,m) when m ≥ 4 is even . . . . . . . . . . . . 18
2-E Comments on the binary cyclic codes C(i,m) . . . . . . . . . . . . . . . . . 21

3 Two families of binary duadic codes with a square-root-like lower bound 22

3-A Known binary duadic codes with a square-root-like lower bound . . . . . 22
3-B The two families of binary duadic codes . . . . . . . . . . . . . . . . . . . 24
3-C Some auxiliary results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3-D Parameters of the two families of duadic codes and their related codes . . 27
3-E Differences among several families of duadic codes . . . . . . . . . . . . . 30

4 Summary and concluding remarks 30

References 31
 3

1. I NTRODUCTION

In this paper, let Fq denote the finite field of order q, where q is a power of a prime p. An
[n, k, d] linear code C over Fq is a k-dimensional subspace of Fnq with minimum (Hamming)
distance d. The dual code of C, denoted by C ⊥ , is defined by

C ⊥ = {b ∈ Fnq : bcT = 0 for all c ∈ C},

where bcT is the standard inner product of two vectors b and c in Fnq .
The linear code C over Fq is said to be cyclic if (c0 , c1 , . . . , cn−1) ∈ C implies (cn−1 , c0 , . . . , cn−2) ∈
C. By identifying each vector (c0 , c1 , . . . , cn−1 ) ∈ Fnq with

c0 + c1 x + c2 x2 + · · · + cn−1 xn−1 ∈ Fq [x]/(xn − 1),

a code C of length n over Fq corresponds to a subset of Fq [x]/(xn − 1). Then C is a cyclic

code if and only if the corresponding subset is an ideal of Fq [x]/(xn − 1). Note that every ideal
of Fq [x]/(xn − 1) is principal. Then there is a monic polynomial g(x) of the smallest degree
such that C = (g(x)) and g(x) | (xn − 1). Then g(x) is called the generator polynomial and
h(x) = (xn − 1)/g(x) is referred to as the check polynomial of C. Throughout this paper, assume
that gcd(q, n) = 1. Denote m = ordn (q), i.e., m is the smallest positive integer such that q m ≡ 1
q m −1
(mod n). Let α be a primitive element of Fqm and put β = α n . Then β is a primitive n-th
root of unity. The set T = {0 ≤ i ≤ n − 1 : g(β i ) = 0} is referred to as the defining set of C
with respect to β. If T contains δ − 1 consecutive integers, then we have the well-known BCH
bound on cyclic codes, i.e., d ≥ δ.
Cyclic codes are interesting in theory, as they are closely related to quite a number of areas
of mathematics such as algebra, algebraic number theory, number theory, combinatorics and
finite geometry. For example, the determination of the weight distributions of irreducible cyclic
codes is the same as the evaluation of certain Gaussian periods [5]. Cyclic codes are important
in practice due to their efficient encoding and decoding algorithms. However, it is theoretically
hard to design cyclic codes of length n with good parameters if n has small divisors more than
1 due to some general theory developed in [16], [17]. This fact is also confirmed by the tables of
best binary cyclic codes in Appendix A.2 of [2]. It is harder to design binary cyclic codes with
good parameters as the alphabet size is too small. It is a much more difficult problem to design
an infinite family of binary cyclic codes such that each code in the family has good parameters.
 4

It is an interesting problem to design an infinite family of binary cyclic codes with good
parameters such that their duals also have good parameters. For convenience, we call such an
infinite family of binary cyclic codes a dually-good infinite family of binary cyclic codes. Dully-
good infinite families of binary cyclic codes with small or large dimensions relative to their
lengths are relatively easy to construct. The binary Hamming codes and the punctured binary
second-order Reed-Muller codes are two dully-good infinite families of binary cyclic codes with
a large and small dimension, respectively. However, only a small number of dully-good infinite
families of binary cyclic codes with parameters [n, k] and (n − 6)/3 ≤ k ≤ 2(n + 6)/3 are
known in the literature. The following is a list of such binary cyclic codes known to the authors.
1) The family of binary quadratic-residue codes.
2) The punctured binary Reed-Muller codes of length 2m − 1 and order (m − 1)/2, where m
is odd.
3) Two families of cyclic codes presented in [14].
It is in general very hard to determine the minimum distance of a cyclic code with parameters
[n, k] and (n − 6)/3 ≤ k ≤ 2(n + 6)/3 as the dimension k is neither small nor large compared
with a large length n [1], [11], [12]. If it is impossible to determine the minimum distance of
such a code, the best one can do is to develop a good lower bound on the minimum distance of
the code. This is the only way to show that such a code has a good error-correcting capability.
However, this is not easy either. It is more difficult to develop good lower bounds on the minimum
distances of both C and C ⊥ [7], [15]. This explains why it is very difficult to find a dully-good
infinite family of binary cyclic codes with parameters [n, k] and (n − 6)/3 ≤ k ≤ 2(n + 6)/3.
Inspired by the works in [14], our objectives in this paper are the construction and analyses of
five dully-good infinite families of binary cyclic codes with parameters [n, k] and (n − 6)/3 ≤
k ≤ 2(n + 6)/3. Three of the five infinite families of binary cyclic codes and their duals have
a very good lower bound on their minimum distances and contain distance-optimal codes. The
other two families of binary cyclic codes are composed of binary duadic codes with a square-
root-like lower bound on their minimum distances. As a by-product, two families of self-dual
binary codes with a square-root-like lower bound on their minimum distances are obtained.
In this paper by the Database we mean the tables of best known linear codes [8], which are
maintained by Markus Grassl at http://www.codetables.de/. We inform the reader that all the
code examples given in this paper are computed by the Magma software package.
The rest of this paper is organised as follows. Section 2 introduces and analyses the first three
 5

families of binary cyclic codes and their related codes. Section 3 constructs and analyses the
two families of binary duadic codes and their related codes. Section 4 concludes this paper and
proposes some open problems.

2. T HE FIRST THREE FAMILIES OF BINARY CYCLIC CODES AND THEIR DUALS

A. The construction of the first three families of binary cyclic codes

Let m ≥ 2 be a positive integer and let n = 2m − 1. Let Zn = {0, 1, 2, . . . , n − 1} be the ring

of integers modulo n. For any s ∈ Zn , the 2-cyclotomic coset of s modulo n is defined by

Cs(2,n) = {s, s2, s22, . . . , s2ls−1 } mod n ⊆ Zn ,

where ls is the smallest positive integer such that s ≡ s2ls (mod n). For an integer i with
0 ≤ i ≤ 2m − 1, let
i = im−1 2m−1 + im−2 2m−2 + · · · + i1 2 + i0

be the 2-adic expansion of i, where ij ∈ {0, 1} for 0 ≤ j ≤ m − 1. We will also write

i = (im−1 , im−2 , . . . , i1 , i0 ) and call it the 2-adic expansion of i in the sequel. For any i with
m−1
ij . Let α be a generator of F∗2m . We define a polynomial
P
0 ≤ i ≤ n − 1, define w2 (i) =
j=0
Y
g(i,m) (x) = (x − αj ) (1)
1≤i≤n−1
w2 (j)≡i (mod 3)

for each i ∈ {0, 1, 2}. Note that w2 (a1 ) = w2 (a2 ) if a1 and a2 belong to the same cyclotomic
coset. It then follows that g(i,m) (x) ∈ F2 [x]. Let C(i,m) denote the binary cyclic code of length
n = 2m − 1 with generator polynomial g(i,m) (x) for i = 0, 1, 2. When m = 3 and 4, Table I
shows that the three families of binary cyclic codes and their dual codes contain some optimal
binary cyclic codes, which motivated us to study the parameters of the three families of binary
cyclic codes C(i,m) and their dual codes.
In the following three subsections, the dimensions of the three families of binary cyclic codes
C(i,m) and their dual codes will be determined and lower bounds on their minimum distances
will be developed with the BCH bound on cyclic codes.

B. Some auxiliary results

In this subsection, we will present some auxiliary results on the defining sets of the three
families of binary cyclic codes C(i,m) , which will play an important role in developing good
 6

TABLE I
PARAMETERS OF C(i,m) FOR m = 3, 4

Code Parameters Optimality Code Parameters Optimality

C(0,3) [7, 7, 1] Optimal C(0,4) [15, 11, 3] Optimal
C(1,3) [7, 4, 3] Optimal C(1,4) [15, 11, 3] Optimal
C(2,3) [7, 4, 3] Optimal C(2,4) [15, 9, 4] Optimal
⊥ ⊥
C(0,3) [7, 0] / C(0,4) [15, 4, 8] Optimal
⊥ ⊥
C(1,3) [7, 3, 4] Optimal C(1,4) [15, 4, 8] Optimal
⊥ ⊥
C(2,3) [7, 3, 4] Optimal C(2,4) [15, 6, 6] Optimal

lower bounds on the minimum distance of the binary codes. The following well-known lemma
will be employed later.

Lemma 2.1. Let l and m be two positive integers. Then

gcd(am − 1, al − 1) = agcd(m,l) − 1,

where a ≥ 2 is a positive integer.

Below we always assume that n = 2m − 1. Denote

T(i,m) = {1 ≤ j ≤ n − 1 : w2 (j) ≡ i (mod 3)}

for i = 0, 1, 2. By the definition of C(i,m) given by (1), T(i,m) is the defining set of C(i,m) with
respect to the n-th primitive root of unity α.
1) The odd m case:

Lemma 2.2. Let m ≡ 1 (mod 6) ≥ 7. Then we have the following.

1) If v = 2(m−1)/2 − 1, then gcd(v, n) = 1 and

{av : 1 ≤ a ≤ 2(m−1)/2 + 2} ⊆ T(0,m) .

2) If v = 2(m+1)/2 − 1, then gcd(v, n) = 1 and

{av : 1 ≤ a ≤ 2(m−1)/2 + 2} ⊆ T(1,m) .

3) If v = 2(m−1)/2 − 1, then gcd(v, n) = 1 and

{av mod n : 2m−1 + 2(m−1)/2 + 1 ≤ a ≤ 2m−1 + 2(m−1)/2 + 2(m−3)/2 + 1} ⊆ T(2,m) .

 7

Proof. If v = 2(m−1)/2 −1, it follows from Lemma 2.1 that gcd(v, n) = 1. When a = 2(m−1)/2 +2,
we have
av = 2m−1 + 2(m−1)/2 − 2 = 2(2m−2 + 2(m−3)/2 − 1).

Consequently, w2 (av) = (m − 1)/2 ≡ 0 (mod 3). When a = 2(m−1)/2 + 1, av = 2m−1 − 1

and w2 (av) = (m − 1) ≡ 0 (mod 3). When a = 2(m−1)/2 , w2 (av) = w2 (v) = (m − 1)/2 ≡ 0
(mod 3). Now we assume that 1 ≤ a ≤ 2(m−1)/2 − 1. Let a = 2l ā, where ā is odd and l ≥ 0 is
an integer. Then we have 1 ≤ a ≤ 2(m−1)/2 − 1 and the 2-adic expansion of ā given by
(m−3)/2
X
ā = ai 2i .
i=0

Since ā is odd, a0 = 1. We have

(m−3)/2 (m−3)/2
X X
i+(m−1)/2
āv = ai 2 + (1 − ai )2i + 1.
i=1 i=0

It then follows that

m−1 m−1
w2 (āv) = w2 (ā) − 1 + 1 + − w2 (ā) = ≡ 0 (mod 3).
2 2
The desired conclusion in the first case then follows.
If v = 2(m+1)/2 − 1, it follows from Lemma 2.1 that

gcd(v, n) = 2gcd((m+1)/2,m) − 1 = 2gcd((m+1)/2,(m−1)/2) − 1 = 1.

When a = 2(m−1)/2 , it is easy to see that

m+1
w2 (av) = w2 (v) = ≡ 1 (mod 3).
2
Furthermore, one can similarly check that w2 (av) ≡ 1 (mod 3) for a = 2(m−1)/2 + 1 and
2(m−1)/2 + 2. Next, we assume that 1 ≤ a ≤ 2(m−1)/2 − 1. Let a = 2l ā, where ā is odd and l ≥ 0
is an integer. Then we have 1 ≤ ā ≤ 2(m−1)/2 − 1. Let the 2-adic expansion of ā be given by
(m−3)/2
X
ā = ai 2i .
i=0

Since ā is odd, a0 = 1. Then

āv = ā2(m+1)/2 − ā
(m−3)/2 (m−3)/2
X X
i+(m+1)/2 (m−1)/2
= ai 2 +2 + (1 − ai )2i + 1.
i=1 i=0
 8

As a result, we have
m−1 m+1
w2 (āv) = w2 (ā) − 1 + 2 + − w2 (ā) = ≡ 1 (mod 3).
2 2
If v = 2(m−1)/2 − 1, it follows from Lemma 2.1 that gcd(v, n) = 1. When a = 2m−1 +
2(m−1)/2 + 1, one can check that

w2 (av) = w2 (v) = (m − 3)/2 ≡ 2 (mod 3).

(m−5)/2
Write a = 2m−1 + 2(m−1)/2 + 1 + t for 1 ≤ t ≤ 2(m−3)/2 − 1 and t = ti 2i . Then we have
P
i=0

(m−5)/2
X
m−1 (m−1)/2
av = (2 +2 +1+ ti 2i )(2(m−1)/2 − 1)
i=0
(m−5)/2 (m−5)/2
X X
(m−1)/2 m−1 i+(m−1)/2
=2 2 −1+ ti 2 − ti 2i
i=0 i=0
(m−5)/2 (m−5)/2
X X
≡ ti 2i+(m−1)/2 + (1 − ti )2i (mod n).
i=0 i=0

It then follows that

m−3 m−3
w2 (av mod n) = w2 (t) + − w2 (t) = ≡2 (mod 3).
2 2
This completes the proof.

Lemma 2.3. Let m ≡ 3 (mod 6) ≥ 9. Then we have the following.

1) If v = 2(m−1)/2 − 1, then gcd(v, n) = 1 and

{av mod n : 2m−1 + 2(m−1)/2 + 1 ≤ a ≤ 2m−1 + 2(m−1)/2 + 2(m−3)/2 } ⊆ T(0,m) .

2) If v = 2(m−1)/2 − 1, then gcd(v, n) = 1 and

{av : 1 ≤ a ≤ 2(m−1)/2 } ⊆ T(1,m) .

3) If v = 2(m+1)/2 − 1, then gcd(v, n) = 1 and

{av : 1 ≤ a ≤ 2(m−1)/2 } ⊆ T(2,m) .

Proof. The proof is very similar to that of Lemma 2.2 and omitted here.

Lemma 2.4. Let m ≡ 5 (mod 6) ≥ 5. Then we have the following.

 9

1) If v = 2(m+1)/2 − 1, then gcd(v, n) = 1 and

{av : 1 ≤ a ≤ 2(m−1)/2 } ⊆ T(0,m) .

2) If v = 2(m+1)/2 − 1, then gcd(v, n) = 1 and

{av mod n : 2m−1 + 2(m−1)/2 + 1 ≤ a ≤ 2m−1 + 2(m−1)/2 + 2(m−3)/2 } ⊆ T(1,m) .

3) If v = 2(m−1)/2 − 1, then gcd(v, n) = 1 and

{av : 1 ≤ a ≤ 2(m−1)/2 } ⊆ T(2,m) .

Proof. The proof is very similar to that of Lemma 2.2 and omitted here.

2) The even m case:

Lemma 2.5. Let m ≡ 0 (mod 4) ≥ 4 and v = 2(m−2)/2 − 1. Then gcd(v, n) = 1 and





 T(2,m) , if m ≡ 0 (mod 6);

{av : 1 ≤ a ≤ 2(m−2)/2 } ⊆ T(0,m) , if m ≡ 2 (mod 6);



if m ≡ 4

T
(1,m) , (mod 6).

Proof. It follows from Lemma 2.1 that

gcd(v, n) = 2gcd((m−2)/2,m) − 1 = 2gcd((m−2)/2,(m+2)/2) − 1 = 2gcd((m−2)/2,2) − 1 = 1.

We begin to prove the second conclusion. When a = 2(m−2)/2 , we have


2 (mod 3), if m ≡ 0 (mod 6);



m−2 
w2 (av) = w2 (v) = ≡ 0 (mod 3), if m ≡ 2 (mod 6);
2 


1 (mod 3), if m ≡ 4 (mod 6).


Now assume that 1 ≤ a ≤ 2(m−2)/2 − 1. Let a = 2l ā, where ā is odd and l ≥ 0 is an integer.
Let the 2-adic expansion of ā be given by
(m−4)/2
X
ā = ai 2i ,
i=0

where a0 = 1. We have
(m−4)/2 (m−4)/2
X X
i+(m−2)/2
āv = ai 2 + (1 − ai )2i + 1.
i=1 i=0
 10

It then follows that

m−2
w2 (av) = w2 (āv) = w2 (ā) − 1 + 1 + − w2 (ā)
2




 2 (mod 3), if m ≡ 0 (mod 6);
m−2 
= ≡ 0 (mod 3), if m ≡ 2 (mod 6);
2 


1 (mod 3), if m ≡ 4 (mod 6).


This completes the proof.

Lemma 2.6. Let m ≡ 2 (mod 4) ≥ 4 and v = 2(m−4)/2 − 1. Then gcd(v, n) = 1 and





 T(1,m) , if m ≡ 0 (mod 6);

{av : 1 ≤ a ≤ 2(m−4)/2 } ⊆ T(2,m) , if m ≡ 2 (mod 6);



T
(0,m) , if m ≡ 4 (mod 6).


Proof. It follows from Lemma 2.1 that

gcd(v, n) = 2gcd((m−4)/2,m) − 1 = 2gcd((m−4)/2,(m+4)/2) − 1 = 1.

When a = 2(m−4)/2 , we have





1 (mod 3), if m ≡ 0 (mod 6);
m−4 
w2 (av) = w2 (v) = ≡ 2 (mod 3), if m ≡ 2 (mod 6);
2 


0 (mod 3), if m ≡ 4 (mod 6).


Next, we assume that 1 ≤ a ≤ 2(m−4)/2 − 1. Let a = 2l ā, where ā is odd and l ≥ 0 is an integer.
Let the 2-adic expansion of ā be given by
(m−6)/2
X
ā = ai 2i ,
i=0
where a0 = 1. One can check that
(m−6)/2 (m−6)/2
X X
i+(m−2)/2
āv = ai 2 + (1 − ai )2i + 1.
i=1 i=0
It then follows that

1 (mod 3),


 if m ≡ 0 (mod 6);
m−4 
w2 (av) = w2 (āv) = ≡ 2 (mod 3), if m ≡ 2 (mod 6);
2 


0 (mod 3), if m ≡ 4 (mod 6).


Summarizing the discussions above completes proof.

 11

C. Parameters of the binary codes C(i,m)

In this subsection, we investigate the dimensions and minimum distances of the binary cyclic
codes C(i,m) for i = 0, 1, 2.
√
2π −1
Let ω = e 3 ∈ C be a 3-th primitive root of unity, i.e., ω 2 + ω + 1 = 0. Denote
     
X m X m X m
s0 = , s1 = , and s2 = . (2)
0≤i≤m
i 0≤i≤m
i 0≤i≤m
i
i≡0 (mod 3) i≡1 (mod 3) i≡2 (mod 3)

Then we have

(1 + ω)m = s0 + s1 ω + s2 ω 2 , (1 + ω 2 )m = s0 + s1 ω 2 + s2 ω.

It follows from (1 + ω)m (1 + ω 2 )m = 1 that

s20 + s21 + s22 − s0 s1 − s0 s2 − s1 s2 = 1. (3)

1) Parameters of C(i,m) when m ≥ 5 is odd: When m ≡ 1 (mod 6) ≥ 7 is odd, the parameters

of C(i,m) for i = 0, 1, 2 are investigated in the following theorem.

Theorem 2.7. Let m ≡ 1 (mod 6) ≥ 7 be an odd integer. Then the codes C(0,m) and C(1,m) have
parameters
[2m − 1, (2m+1 − 1)/3, d ≥ 2(m−1)/2 + 3],

and the code C(2,m) has parameters

[2m − 1, (2m+1 − 1)/3, d ≥ 2(m−3)/2 + 2].

Proof. Note that m ≡ 1 (mod 6) is odd and w2 (i)+w2 (n−i) = m for each i with 1 ≤ i ≤ n−1.
Hence i ∈ T(0,m) if and only if n − i ∈ T(1,m) , which means that

|T(0,m) | = |T(1,m) |.

In this case, by (2), we have

s0 = |T(0,m) | + 1, s1 = |T(1,m) | + 1, s2 = |T(2,m) |.

One can compute from (3) that

(|T(0,m) | + 1 − |T(2,m) |)2 = 1.

Then
|T(0,m) | = |T(2,m) | or |T(2,m) | = |T(0,m) | + 2.
 12

We assert that |T(0,m) | = |T(2,m) | holds, otherwise, we have |T(2,m) | = |T(0,m) | + 2 and 3|T(0,m) | +
2 = 2m − 2, which leads to a contradiction since |T(0,m) | = (2m − 4)/3 is not a integer.
Consequently,
|T(0,m) | = |T(1,m) | = |T(2,m) | = (2m − 2)/3

and
dim(C(0,m) ) = dim(C(1,m) ) = dim(C(2,m) ) = n − (2m − 2)/3 = (2m+1 − 1)/3.
m−1
Denote v = 2 2 − 1. It follows from Lemma 2.2 that gcd(v, n) = 1. Let v be the integer
satisfying vv ≡ 1 (mod n). Write γ = αv . It is deduced from Lemma 2.2 that the defining set of
C(0,m) with respect to γ contains the set 1, 2, ..., 2(m−1)/2 + 2 . The lower bound on the minimum


distance of C(0,m) then follows from the BCH bound on cyclic codes. The desired conclusion
on the minimum distances of C(1,m) and C(2,m) can be similarly obtained. This completes the
proof.

When m ≡ 3 (mod 6) ≥ 9 is odd, the parameters of C(i,m) for i = 0, 1, 2 are treated in the
following theorem.

Theorem 2.8. Let m ≡ 3 (mod 6) ≥ 9 be an odd integer. Then the codes C(1,m) and C(2,m) have
parameters
[2m − 1, (2m+1 − 4)/3, d ≥ 2(m−1)/2 + 1],

and the code C(0,m) has parameters

[2m − 1, (2m+1 + 5)/3, d ≥ 2(m−3)/2 + 1].

Proof. Note that m ≡ 3 (mod 6) and w2 (i) + w2 (n − i) = m for each i with 1 ≤ i ≤ n − 1.

Hence i ∈ T(1,m) if and only if n − i ∈ T(2,m) . Then

|T(1,m) | = |T(2,m) |.

In this case, by (2), we have

s0 = |T(0,m) | + 2, s1 = |T(1,m) |, s2 = |T(2,m) |.

One can similarly obtain from (3) that |T(1,m) | = |T(0,m) | + 3. Thus, we have

|T(0,m) | = (2m − 8)/3, |T(1,m) | = |T(2,m) | = (2m + 1)/3.

 13

The desired conclusion on the dimensions then follows. The lower bounds on the minimum
distances can be derived by employing Lemma 2.3. The proof is very similar to that of Theorem
2.7 and omitted.

When m ≡ 5 (mod 6) ≥ 5 is odd, the parameters of C(i,m) for i = 0, 1, 2 are investigated in

the following theorem.

Theorem 2.9. Let m ≡ 5 (mod 6) ≥ 5 be an odd integer. Then the codes C(0,m) and C(2,m) have
parameters
[2m − 1, (2m+1 − 1)/3, d ≥ 2(m−1)/2 + 1],

and the code C(1,m) has parameters

[2m − 1, (2m+1 − 1)/3, d ≥ 2(m−3)/2 + 1].

Proof. Note that m ≡ 5 (mod 6). Then one similarly has

|T(2,m) | = |T(0,m) |

and
s0 = |T(0,m) | + 1, s1 = |T(1,m) |, s2 = |T(2,m) | + 1.

The remainder of the proof is very similar to that of Theorem 2.7 with the help of Lemma 2.4,
and we omit the details.

2) Parameters of C(i,m) when m ≥ 4 is even: Assume that m ≥ 4 is an even integer. We

investigate the parameters of C(i,m) in the two cases: m ≡ 0 (mod 4) and m ≡ 2 (mod 4).
When m ≡ 0 (mod 4) ≥ 4 is even, the parameters of C(i,m) for i = 0, 1, 2 are studied in the
following theorem.

Theorem 2.10. Let m ≡ 0 (mod 4) ≥ 4 be an even integer. Then we have the following.
1) If m ≡ 0 (mod 6) ≥ 6, then C(1,m) and C(2,m) have parameters

[2m − 1, (2m+1 − 2)/3, d ≥ 2(m−2)/2 + 1],

and C(0,m) has parameters

[2m − 1, (2m+1 + 1)/3, d ≥ 2(m−2)/2 + 1].

 14

2) If m ≡ 2 (mod 6) ≥ 8, then C(0,m) and C(2,m) have parameters

[2m − 1, (2m+1 + 1)/3, d ≥ 2(m−2)/2 + 1],

and C(1,m) has parameters

[2m − 1, (2m+1 − 5)/3, d ≥ 2(m−2)/2 + 1].

3) If m ≡ 4 (mod 6) ≥ 4, then C(0,m) and C(1,m) have parameters

[2m − 1, (2m+1 + 1)/3, d ≥ 2(m−2)/2 + 1],

and C(2,m) has parameters

[2m − 1, (2m+1 − 5)/3, d ≥ 2(m−2)/2 + 1].

Proof. When m ≡ 0 (mod 6), we similarly have |T(1,m) | = |T(2,m) | and

s0 = |T(0,m) | + 2, s1 = |T(1,m) |, s2 = |T(2,m) |.

It is deduced from (3) that

|T(0,m) | = (2m − 4)/3, |T(1,m) | = |T(2,m) | = (2m − 1)/3.

When m ≡ 2 (mod 6) ≥ 8, we similarly have |T(0,m) | = |T(2,m) | and

s0 = |T(0,m) | + 1, s1 = |T(1,m) |, s2 = |T(2,m) | + 1.

It is deduced from (3) that

|T(0,m) | = |T(2,m) | = (2m − 4)/3, |T(1,m) | = (2m + 2)/3.

When m ≡ 4 (mod 6) ≥ 4, we similarly have |T(0,m) | = |T(1,m) | and

s0 = |T(0,m) | + 1, s1 = |T(1,m) | + 1, s2 = |T(2,m) |.

It is deduced from (3) that

|T(0,m) | = |T(1,m) | = (2m − 4)/3, |T(2,m) | = (2m + 2)/3.

The desired conclusion on the dimension of C(i,m) then follows.

For v = 2(m−2)/2 − 1, it follows from Lemma 2.5 that gcd(v, n) = 1 if m ≡ 0 (mod 4).
Let v be the multiplicative inverse of v modulo n and let γ = αv . It then follows again from
Lemma 2.5 that the defining set of C(i,m) with respect to γ contains the set 1, 2, ..., 2(m−2)/2 .


The desired lower bound on d then follows from the BCH bound on cyclic codes.
 15

When m ≡ 2 (mod 4) ≥ 4 is even, the parameters of C(i,m) for i = 0, 1, 2 are investigated in

the following theorem.

Theorem 2.11. Let m ≡ 2 (mod 4) ≥ 4 be an even integer. Then we have the following.
1) If m ≡ 0 (mod 6) ≥ 6, then C(1,m) and C(2,m) have parameters

[2m − 1, (2m+1 − 2)/3, d ≥ 2(m−4)/2 + 1],

and C(0,m) has parameters

[2m − 1, (2m+1 + 1)/3, d ≥ 2(m−4)/2 + 1].

2) If m ≡ 2 (mod 6) ≥ 8, then C(0,m) and C(2,m) have parameters

[2m − 1, (2m+1 + 1)/3, d ≥ 2(m−4)/2 + 1],

and C(1,m) has parameters

[2m − 1, (2m+1 − 5)/3, d ≥ 2(m−4)/2 + 1].

3) If m ≡ 4 (mod 6) ≥ 4, then C(0,m) and C(1,m) have parameters

[2m − 1, (2m+1 + 1)/3, d ≥ 2(m−4)/2 + 1],

and C(2,m) has parameters

[2m − 1, (2m+1 − 5)/3, d ≥ 2(m−4)/2 + 1].

Proof. The proof is very similar to that of Theorem 2.10 and omitted here.

⊥
D. Parameters of the dual codes C(i,m)
⊥
In this subsection, we investigate the parameters of the dual codes C(i,m) for i = 0, 1, 2. Their
dimensions are explicitly determined and low bounds on the minimum distances of these codes
are developed.
 16

⊥
1) Parameters of C(i,m) when m ≥ 5 is odd: To develop lower bounds on the minimum
⊥
distances of C(i,m) for odd m ≥ 5, we need the following three lemmas, which can be similarly
proved by using the same techniques given in Section 2-B. Below we only state the lemmas and
omit their proofs.

Lemma 2.12. Let m ≡ 1 (mod 6) ≥ 7. Then we have the following.

1) If v = 2(m+1)/2 − 1, then gcd(v, n) = 1 and

{av mod n : 0 ≤ a ≤ 2(m−1)/2 + 2} ⊆ Zn \ T(0,m) .

2) If v = 2(m−1)/2 − 1, then gcd(v, n) = 1 and

{av : 0 ≤ a ≤ 2(m−1)/2 + 2} ⊆ Zn \ T(1,m) .

3) If v = 2(m+1)/2 − 1, then gcd(v, n) = 1 and

{av mod n : 2m − 2(m−1)/2 − 5 ≤ a ≤ n − 1, or 0 ≤ a ≤ 2(m−1)/2 + 4} ⊆ Zn \ T(2,m) .

Lemma 2.13. Let m ≡ 3 (mod 6) ≥ 9. Then we have the following.

1) If v = 2(m−1)/2 − 1, then gcd(v, n) = 1 and

{av mod n : 2m − 2(m−1)/2 − 5 ≤ a ≤ n − 1, or 0 ≤ a ≤ 2(m−1)/2 + 4} ⊆ Zn \ T(0,m) .

2) If v = 2(m+1)/2 − 1, then gcd(v, n) = 1 and

{av : 0 ≤ a ≤ 2(m−1)/2 } ⊆ Zn \ T(1,m) .

3) If v = 2(m−1)/2 − 1, then gcd(v, n) = 1 and

{av : 0 ≤ a ≤ 2(m−1)/2 } ⊆ Zn \ T(2,m) .

Lemma 2.14. Let m ≡ 5 (mod 6) ≥ 5. Then we have the following.

1) If v = 2(m−1)/2 − 1, then gcd(v, n) = 1 and

{av : 0 ≤ a ≤ 2(m−1)/2 + 2} ⊆ Zn \ T(0,m) .

2) If v = 2(m−1)/2 − 1, then gcd(v, n) = 1 and

{av mod n : 2m − 2(m−1)/2 − 1 ≤ a ≤ n − 1, or 0 ≤ a ≤ 2(m−1)/2 } ⊆ Zn \ T(1,m) .

3) If v = 2(m+1)/2 − 1, then gcd(v, n) = 1 and

{av mod n : 0 ≤ a ≤ 2(m−1)/2 + 2} ⊆ Zn \ T(2,m) .

 17

⊥
The dimensions and lower bounds on the minimum distances of the dual codes C(i,m) for
i = 0, 1, 2 are documented in the following theorem.

Theorem 2.15. Let m ≥ 5 be an odd integer.

⊥
1) The code C(0,m) has parameters [2m − 1, k ⊥ , d⊥ ], where k ⊥ and d⊥ are given as follows:

(2m − 2)/3, if m ≡ 1 or 5 (mod 6) ≥ 5;

⊥
k =
(2m − 8)/3, if m ≡ 3 (mod 6) ≥ 9;


and 
2(m−1)/2 + 4,

if m ≡ 1 or 5 (mod 6) ≥ 5;
d⊥ ≥
2(m+1)/2 + 10,

if m ≡ 3 (mod 6) ≥ 9.
⊥
2) The code C(1,m) has parameters [2m − 1, k ⊥ , d⊥ ], where k ⊥ and d⊥ are given as follows:

(2m − 2)/3, if m ≡ 1 or 5 (mod 6) ≥ 5;

⊥
k =
(2m + 1)/3, if m ≡ 3 (mod 6) ≥ 9;


and 



2(m−1)/2 + 4, if m ≡ 1 (mod 6) ≥ 7;

d⊥ ≥ 2(m−1)/2 + 2, if m ≡ 3 (mod 6) ≥ 9;



2(m+1)/2 + 2,

if m ≡ 5 (mod 6) ≥ 5.
⊥
3) The code C(2,m) has parameters [2m − 1, k ⊥ , d⊥ ], where k ⊥ and d⊥ are given as follows:

(2m − 2)/3, if m ≡ 1 or 5 (mod 6) ≥ 5;

⊥
k =
(2m + 1)/3, if m ≡ 3 (mod 6) ≥ 9;


and 
(m+1)/2
2


 + 10, if m ≡ 1 (mod 6) ≥ 7;

d⊥ ≥ 2(m−1)/2 + 2, if m ≡ 3 (mod 6) ≥ 9;



2(m−1)/2 + 4, if m ≡ 5 (mod 6) ≥ 5.


⊥ ⊥
Proof. We prove the desired conclusion only for the code C(0,m) , as the conclusion for C(1,m)
⊥
and C(2,m) can be similarly proved.
 18

Note that w2 (i) + w2 (n − i) = m for each i with 1 ≤ i ≤ n − 1. Then





 T(1,m) , if m ≡ 1 (mod 6) ≥ 7;

−1
T(0,m) = T(0,m) , if m ≡ 3 (mod 6) ≥ 9; (4)



T
(2,m) , m ≡ 5 (mod 6) ≥ 5.

if
⊥ ⊥
Let T(0,m) be the defining set of C(0,m) . Thus we have

⊥ ⊥ −1 −1
dim(C(0,m) ) = n − |T(0,m) | = n − |Zn \ T(0,m) | = |T(0,m) |.

⊥
The desired conclusion on the dimension of C(0,m) then follows from (4) and Theorems 2.7, 2.8,
⊥
and 2.9. The desired lower bound on the minimum distance of C(0,m) is deduced from the BCH
bound with the help of Lemmas 2.12, 2.13, and 2.14.

Example 1. Let m = 5 and let α be a generator of F∗25 with α5 + α2 + 1 = 0.

⊥
• The codes C(0,5) and C(0,5) have parameters [31, 21, 5] and [31, 10, 12], respectively, and are
optimal according to the Database [8].
⊥
• The codes C(1,5) and C(1,5) have parameters [31, 21, 5] and [31, 10, 10], respectively, and the
former code is optimal according to the Database [8].
⊥
• The codes C(2,5) and C(2,5) have parameters [31, 21, 5] and [31, 10, 12], respectively, and are
optimal according to the Database [8].

⊥
2) Parameters of C(i,m) when m ≥ 4 is even: Assume that m ≥ 4 is an even integer. To develop
⊥
lower bounds on the minimum distances of C(i,m) in this case, the following two lemmas will be
employed later. The two lemmas below can be similarly proved by using the same techniques
given in Section 2-B. Below we only state the lemmas and omit their proofs.

Lemma 2.16. Let m ≡ 0 (mod 4) ≥ 4. Then we have the following.

1) If v = 2(m−2)/2 − 1, then gcd(v, n) = 1 and

{av : 0 ≤ a ≤ 2(m−2)/2 } ⊆ Zn \ T(1,m) ,

 if m ≡ 0 (mod 6);


 {av : 0 ≤ a ≤ 2(m−2)/2 + 4} ⊆ Zn \ T(2,m) , if m ≡ 2 (mod 6);


{av : 0 ≤ a ≤ 2(m−2)/2 + 4} ⊆ Z \ T

if m ≡ 4 (mod 6).
n (0,m) ,
 19

2) If v = 2(m−2)/2 − 1, then gcd(v, n) = 1 and


{av mod n : 2m − 2(m−2)/2 − 3 ≤ a ≤ n − 1, or







0 ≤ a ≤ 2(m−2)/2 + 2} ⊆ Zn \ T(0,m) , if m ≡ 0 (mod 6);







{av mod n : 2m − 2(m−2)/2 − 3 ≤ a ≤ n − 1, or


0 ≤ a ≤ 2(m−2)/2 + 2} ⊆ Zn \ T(1,m) , if m ≡ 2 (mod 6);








{av mod n : 2m − 2(m−2)/2 − 1 ≤ a ≤ n − 1, or







0 ≤ a ≤ 2(m−2)/2 } ⊆ Zn \ T(2,m) ,

if m ≡ 4 (mod 6).
3) If v = 2(m+2)/2 − 1, then gcd(v, n) = 1 and




 {av mod n : 0 ≤ a ≤ 2(m−2)/2 } ⊆ Zn \ T(2,m) , if m ≡ 0 (mod 6);

 {av mod n : 0 ≤ a ≤ 2(m−2)/2 + 4} ⊆ Zn \ T(0,m) , if m ≡ 2 (mod 6);


{av mod n : 0 ≤ a ≤ 2(m−2)/2 + 4} ⊆ Z \ T

if m ≡ 4
n (1,m) , (mod 6).

Lemma 2.17. Let m ≡ 2 (mod 4) ≥ 6. Then we have the following.

1) If v = 2(m−4)/2 − 1, then gcd(v, n) = 1 and




 {av : 0 ≤ a ≤ 2(m−4)/2 } ⊆ Zn \ T(2,m) , if m ≡ 0 (mod 6);

 {av : 0 ≤ a ≤ 2(m−4)/2 + 2} ⊆ Zn \ T(0,m) , if m ≡ 2 (mod 6);


{av : 0 ≤ a ≤ 2(m−4)/2 + 2} ⊆ Zn \ T(1,m) , if m ≡ 4 (mod 6).



2) If v = 2(m−4)/2 − 1, then gcd(v, n) = 1 and


{av mod n : 2m − 2(m−4)/2 − 5 ≤ a ≤ n − 1, or







0 ≤ a ≤ 2(m−4)/2 + 4} ⊆ Zn \ T(0,m) , if m ≡ 0 (mod 6);







{av mod n : 2m − 2(m−4)/2 − 1 ≤ a ≤ n − 1, or


0 ≤ a ≤ 2(m−4)/2 } ⊆ Zn \ T(1,m) , if m ≡ 2 (mod 6);








{av mod n : 2m − 2(m−4)/2 − 5 ≤ a ≤ n − 1, or







0 ≤ a ≤ 2(m−4)/2 + 4} ⊆ Zn \ T(2,m) ,

if m ≡ 4 (mod 6).
3) If v = 2(m+4)/2 − 1, then gcd(v, n) = 1 and




 {av mod n : 0 ≤ a ≤ 2(m−4)/2 } ⊆ Zn \ T(1,m) , if m ≡ 0 (mod 6);

 {av mod n : 0 ≤ a ≤ 2(m−4)/2 + 2} ⊆ Zn \ T(2,m) , if m ≡ 2 (mod 6);


{av mod n : 0 ≤ a ≤ 2(m−4)/2 + 2} ⊆ Zn \ T(0,m) , if m ≡ 4

 (mod 6).
 20

⊥
When m ≥ 4 is even, the parameters of C(i,m) for i = 0, 1, 2 are treated in the following
theorem.

Theorem 2.18. Let m ≥ 4 be even. Then we have the following.

⊥
1) The code C(0,m) has parameters [2m − 1, (2m − 4)/3, d⊥ ], where




 2m/2 + 6, if m ≡ 0 (mod 6);

d⊥ ≥ 2(m−2)/2 + 6, if m ≡ 2 (mod 6);



2(m−2)/2 + 6, if m ≡ 4 (mod 6),


when m ≡ 0 (mod 4), or





2(m−2)/2 + 10, if m ≡ 0 (mod 6);

d⊥ ≥ 2(m−4)/2 + 4, if m ≡ 2 (mod 6);



2(m−4)/2 + 4,

if m ≡ 4 (mod 6),

when m ≡ 2 (mod 4).

⊥
2) The code C(1,m) has parameters [2m − 1, k ⊥ , d⊥ ], where




 (2m − 1)/3, if m ≡ 0 (mod 6);

k ⊥ = (2m + 2)/3, if m ≡ 2 (mod 6);



(2m − 4)/3, if m ≡ 4 (mod 6),


and 



 2(m−2)/2 + 2, if m ≡ 0 (mod 6);

d⊥ ≥ 2m/2 + 6, if m ≡ 2 (mod 6);



2(m−2)/2 + 6,

if m ≡ 4 (mod 6),

when m ≡ 0 (mod 4), or





 2(m−4)/2 + 2, if m ≡ 0 (mod 6);

d⊥ ≥ 2(m−2)/2 + 2, if m ≡ 2 (mod 6);



2(m−4)/2 + 4,

if m ≡ 4 (mod 6),

when m ≡ 2 (mod 4).

 21

⊥
3) The code C(2,m) has parameters [2m − 1, k ⊥ , d⊥ ], where




 (2m − 1)/3, if m ≡ 0 (mod 6);

k ⊥ = (2m − 4)/3, if m ≡ 2 (mod 6);



(2m + 2)/3, if m ≡ 4 (mod 6),


and 



 2(m−2)/2 + 2, if m ≡ 0 (mod 6);

d⊥ ≥ 2(m−2)/2 + 6, if m ≡ 2 (mod 6);



2m/2 + 2,

if m ≡ 4 (mod 6),

when m ≡ 0 (mod 4), or


2(m−4)/2 + 2,

 if m ≡ 0 (mod 6);


⊥
d ≥ 2(m−4)/2 + 4, if m ≡ 2 (mod 6);



2(m−2)/2 + 10,

if m ≡ 4 (mod 6),

when m ≡ 2 (mod 4).

Proof. The proof is very similar to that of Theorem 2.15 and omitted here.

Example 2. Let m = 6 and let α be a generator of F∗26 with α6 + α4 + α3 + α + 1 = 0.

⊥
• The codes C(0,6) and C(0,6) have parameters [63, 43, 6] and [63, 20, 14], respectively.
⊥
• The codes C(1,6) and C(1,6) have parameters [63, 42, 6] and [63, 21, 16], respectively.
⊥
• The codes C(2,6) and C(2,6) have parameters [63, 42, 6] and [63, 21, 16], respectively.

E. Comments on the binary cyclic codes C(i,m)

The dimensions dim(C(i,m) ) of these codes are around 2n/3 and the dimensions of their duals
are around n/3. Experimental data shows that the lower bounds on the minimum distances of
⊥
C(i,m) and C(i,m) developed in the previous subsections are very good. In particular, the three
families of codes contain distance-optimal codes (see Example 1). Hence, they are three dully-
good infinite families of binary cyclic codes.
 22

3. T WO FAMILIES OF BINARY DUADIC CODES WITH A SQUARE - ROOT- LIKE LOWER BOUND

A. Known binary duadic codes with a square-root-like lower bound

In this subsection, we introduce binary duadic codes and survey binary duadic codes with a
square-root-like lower bound on their minimum distances, which are binary duadic codes with
√
parameters of the form [n, (n ± 1)/2, d] such that d is very close to n.
Let n be an odd positive integer and let Zn denote the ring of integers modulo n. Let m =
ordn (2), i.e., the order of 2 modulo n. Let β be an n-th primitive root of unity in F2m . Let S1
and S2 be two subsets of Zn such that
• S1 ∩ S2 = ∅ and S1 ∪ S2 = Zn \ {0}, and
• both S1 and S2 are the union of some 2-cyclotomic cosets modulo n.
If there is a unit µ ∈ Zn such that S1 µ = S2 and S2 µ = S1 , then (S1 , S2 , µ) is called a splitting
of Zn .
Let (S1 , S2 , µ) be a splitting of Zn . Define
Y
gi (x) = (x − β i ) and g̃i (x) = (x − 1)gi (x)
i∈Si

for i ∈ {1, 2}. The pair of cyclic codes C1 and C2 of length n over F2 with generator polynomials
g1 (x) and g2 (x) are called odd-like duadic codes, and the pair of cyclic codes C˜1 and C˜2 of length
n over F2 with generator polynomials g̃1 (x) and g̃2 (x) are called even-like duadic codes.
By definition, C1 and C2 have parameters [n, (n+1)/2] and C˜1 and C˜2 have parameters [n, (n−
1)/2]. For odd-like duadic codes, we have the following result [6, Theorem 6.5.2].

Theorem 3.1 (Square root bound). Let C1 and C2 be a pair of odd-like duadic codes of length
n over F2 . Let do be their (common) minimum odd weight. Then the following hold:
1) d2o ≥ n.
2) If the splitting defining the duadic codes is given by µ = −1, then d2o − do + 1 ≥ n.
3) Suppose d2o − do + 1 = n, where do > 2, and assume that the splitting defining the duadic
codes is given by µ = −1. Then do is the minimum weight of both C1 and C2 .

Binary duadic codes are theoretically attractive due to the following facts [4]:
• All code examples demonstrate that duadic codes of prime lengths have a square-root bound
on their minimum distances. It is known that duadic codes that are not quadratic-residue
codes have a square-root bound on their minimum odd weight (see Theorem 3.1).
 23

• Duadic codes could be the best class of cyclic codes of certain lengths. For example, consider
n = 31 and all binary cyclic codes of length 31 and dimension 16. There are four [31, 16]
binary cyclic codes up to equivalence, one is the quadratic-residue code and the other is
duadic, not quadratic-residue. Both of these have minimum weight 7, the other two codes
have minimum weights 5 and 6 [6, Chapter 6].
• Another example of very good duadic codes that are not quadratic-residue codes are the
[113, 57, 18] codes, which have higher minimum weight than the quadratic residue code of
the same length [13]. Further the [151, 76, 23] and [233, 117, 32] duadic codes have higher
minimum weights than the quadratic residue codes of the same lengths [13] and indeed are
the best codes known of their length.
• Every self-dual extended cyclic binary code is the extended code of a duadic code [13, p.
10].
Duadic codes are a generalisation of the quadratic residue codes. They were introduced and
investigated by Leon, Masley and Pless [9], Leon [10], and Pless, Masley and Leon [13], where
a number of properties are proved. Also all binary duadic codes of length until 241 are described
in [13]. The total number of binary duadic codes of prime power lengths and their constructions
were presented in [3] and [4]. Further information on the existence, constructions, and properties
of duadic codes can be found in [6, Chapter 6].
pm
Q
It is proven that binary duadic codes of length n exist if and only if n = i i
i
where each
pi ≡ ±1 (mod 8) [9]. In general, there are many binary duadic codes of length n (see [3] and
[4]). Experimental computations show that the minimum distance of many binary duadic codes
is poor [6, Chapter 6]. The minimum weight of an odd-like duadic code may be even. It is open
which binary duadic codes have an odd minimum weight. Hence, the lower bound in Theorem
3.1 cannot be used to develop a lower bound on the minimum weight of a binary duadic code.
The only known infinite families of binary duadic codes with a square-root-like lower bound on
their minimum distances are the following:
• Binary quadratic residue codes with parameters [n, (n + 1)/2, d], where d2 ≥ n and n ≡ ±1
(mod 8) is a prime.
• The punctured binary Reed-Muller codes of order (m − 1)/2 which has parameters [2m −
1, 2m−1 , 2(m+1)/2 − 1], where m is odd.
• Two infinite families of binary duadic codes with parameters [2m − 1, 2m−1 , d]] presented
 24

in [14], where m > 3 is odd and d has a square-root-like lower bound.

Motivated by these facts above, we will present two more infinite families of binary duadic
codes with a square-root-like lower bound on their minimum distances in the next subsection.

B. The two families of binary duadic codes

From now on, we fix n = 2m − 1 and let α denote a primitive element of F2m , where m ≥ 3
is odd. Define a polynomial
Y
g(i1 ,i2 ,m) (x) = (x − αj ), (5)
1≤i≤n−1
w2 (j)≡i1 or i2 (mod 4)

where i1 and i2 are a pair of distinct elements in the set {0, 1, 2, 3}. It is easily seen that
g(i1 ,i2 ,m) (x) ∈ F2 [x]. Let C(i1 ,i2 ,m) denote the binary cyclic code of length n = 2m − 1 with
generator polynomial g(i1 ,i2 ,m) (x). Denote

T(i1 ,i2 ,m) = {1 ≤ j ≤ n − 1 : w2 (j) ≡ i1 or i2 (mod 4)}

for i = 0, 1, 2, 3. It is clear that T(i1 ,i2 ,m) is the defining set of C(i1 ,i2 ,m) with respect to the n-th
primitive root of unity α. When m is odd, it is easy to see that C(0,2,m) and C(1,3,m) form a pair
of odd-like duadic codes, which were originally proposed and studied in [14]. We will prove
the following two statements later:
• When m ≡ 1 (mod 4), C(0,3,m) and C(1,2,m) form a pair of odd-like duadic codes.
• When m ≡ 3 (mod 4), C(0,1,m) and C(2,3,m) form a pair of odd-like duadic codes.
In the following subsections, we will investigate the dimensions and minimum distances of the
two families of duadic codes and their dual and extended codes.

C. Some auxiliary results

To develop lower bounds on the minimum distances of the two families of duadic codes, we
need some auxiliary results about their defining sets.

Lemma 3.2. Let m ≡ 1 (mod 8) ≥ 9. Then we have the following.

1) If v = 2(m−1)/2 − 1, then gcd(v, n) = 1 and

{av : 1 ≤ a ≤ 2(m−1)/2 + 2} ⊆ T(0,3,m) .

 25

2) If v = 2(m+1)/2 − 1, then gcd(v, n) = 1 and

{av : 1 ≤ a ≤ 2(m−1)/2 + 2} ⊆ T(1,2,m) .

Proof. If v = 2(m−1)/2 −1, it follows from Lemma 2.1 that gcd(v, n) = 1. When a = 2(m−1)/2 +2,
we have
av = 2m−1 + 2(m−1)/2 − 2 = 2(2m−2 + 2(m−3)/2 − 1).

Consequently, w2 (av) = (m − 1)/2 ≡ 0 (mod 4). When a = 2(m−1)/2 + 1, av = 2m−1 − 1 and

w2 (av) = m − 1 ≡ 0 (mod 4). When a = 2(m−1)/2 , w2 (av) = w2 (v) = (m − 1)/2 ≡ 0 (mod 4).
Now we assume that 1 ≤ a ≤ 2(m−1)/2 − 1. Let a = 2l ā, where ā is odd and l ≥ 0 is an integer.
Then we have 1 ≤ ā ≤ 2(m−1)/2 − 1 and the 2-adic expansion of ā is given by
(m−3)/2
X
ā = ai 2i .
i=0

Since ā is odd, a0 = 1. We have

(m−3)/2 (m−3)/2
X X
i+(m−1)/2
āv = ai 2 + (1 − ai )2i + 1.
i=1 i=0

It then follows that

m−1 m−1
w2 (āv) = w2 (ā) − 1 + 1 + − w2 (ā) = ≡ 0 (mod 4).
2 2
The desired conclusion in the first case then follows.
If v = 2(m+1)/2 − 1, it follows from Lemma 2.1 that

gcd(v, n) = 2gcd((m+1)/2,m) − 1 = 2gcd((m+1)/2,(m−1)/2) − 1 = 1.

When a = 2(m−1)/2 , it is easy to see that

m+1
w2 (av) = w2 (v) = ≡ 1 (mod 4).
2
Furthermore, one can similarly check that w2 (av) ≡ 1 (mod 4) for a = 2(m−1)/2 + 1 and
2(m−1)/2 + 2. Next, we assume that 1 ≤ a ≤ 2(m−1)/2 − 1. Let a = 2l ā, where ā is odd and l ≥ 0
is an integer. Then we have 1 ≤ ā ≤ 2(m−1)/2 − 1. Let the 2-adic expansion of ā be given by
(m−3)/2
X
ā = ai 2i .
i=0
 26

Since ā is odd, a0 = 1. Then

āv = ā2(m+1)/2 − ā
(m−3)/2 (m−3)/2
X X
i+(m+1)/2 (m−1)/2
= ai 2 +2 + (1 − ai )2i + 1.
i=1 i=0

As a result, we have
m−1 m+1
w2 (āv) = w2 (ā) − 1 + 2 + − w2 (ā) = ≡ 1 (mod 4).
2 2
This completes the proof.

The following three lemmas can be similarly proved and their proofs are omitted here.

Lemma 3.3. Let m ≡ 3 (mod 8) ≥ 3. Then we have the following.

1) If v = 2(m−1)/2 − 1, then gcd(v, n) = 1 and

{av : 1 ≤ a ≤ 2(m−1)/2 } ⊆ T(0,1,m) .

2) If v = 2(m+1)/2 − 1, then gcd(v, n) = 1 and

{av : 1 ≤ a ≤ 2(m−1)/2 } ⊆ T(2,3,m) .

Lemma 3.4. Let m ≡ 5 (mod 8) ≥ 5. Then we have the following.

1) If v = 2(m+1)/2 − 1, then gcd(v, n) = 1 and

{av : 1 ≤ a ≤ 2(m−1)/2 } ⊆ T(0,3,m) .

2) If v = 2(m−1)/2 − 1, then gcd(v, n) = 1 and

{av : 1 ≤ a ≤ 2(m−1)/2 } ⊆ T(1,2,m) .

Lemma 3.5. Let m ≡ 7 (mod 8) ≥ 7. Then we have the following.

1) If v = 2(m+1)/2 − 1, then gcd(v, n) = 1 and

{av : 1 ≤ a ≤ 2(m−1)/2 + 2} ⊆ T(0,1,m) .

2) If v = 2(m−1)/2 − 1, then gcd(v, n) = 1 and

{av : 1 ≤ a ≤ 2(m−1)/2 + 2} ⊆ T(2,3,m) .

 27

D. Parameters of the two families of duadic codes and their related codes

When m ≡ 1 (mod 4) ≥ 5, the parameters of C(0,3,m) and C(1,2,m) are treated in the following
theorem.

Theorem 3.6. Let m ≡ 1 (mod 4) ≥ 5 be an integer. Then C(0,3,m) and C(1,2,m) form a pair of
odd-like duadic codes with parameters [2m − 1, 2m−1 , d], where

 2(m−1)/2 + 3 if m ≡ 1 (mod 8),
d≥
 2(m−1)/2 + 1 if m ≡ 5 (mod 8).

Proof. Note that m ≡ 1 (mod 4) and w2 (i) = m − w2 (n − i) for each i with 1 ≤ i ≤ n − 1.

Hence, i ∈ T(0,3,m) if and only if n−i ∈ T(1,2,m) . It then follows that T(0,3,m) and T(1,2,m) partition
Zn \ {0} and
T(0,3,m) = −T(1,2,m) and T(1,2,m) = −T(0,3,m) .

It then follows that C(0,3,m) and C(1,2,m) form a pair of duadic codes with length n and dimension
(n + 1)/2. Hence, the two codes have the same minimum distance d.
We only prove the lower bounds on minimum distannce d in the case that m ≡ 1 (mod 8) as
m−1
it is similar to prove the desired conclusion for m ≡ 5 (mod 8). Denote v = 2 2 −1. It follows
from Lemma 3.2 that gcd(v, n) = 1. Let v be the integer satisfying vv ≡ 1 (mod n). Write
γ = αv . It is deduced from Lemma 3.2 that defining set of C(0,3,m) with respect to γ contains
the set 1, 2, ..., 2(m−1)/2 + 2 . The desired lower bound on the minimum distance of C(0,3,m)


then follows from the BCH bound on cyclic codes. The desired conclusion on the minimum
distance of C(1,2,m) follows naturally, as the two duadic codes have the same minimum distance.
This completes the proof.
⊥ ⊥
When m ≡ 1 (mod 4) ≥ 5, the parameters of the dual codes C(0,3,m) and C(1,2,m) are studied
in the following theorem.

⊥ ⊥
Theorem 3.7. Let m ≡ 1 (mod 4) ≥ 5 be an integer. Then C(0,3,m) and C(1,2,m) form a pair of
even-like duadic codes with parameters [2m − 1, 2m−1 − 1, d⊥ ], where

 2(m−1)/2 + 4 if m ≡ 1 (mod 8),
⊥
d ≥
 2(m−1)/2 + 2 if m ≡ 5 (mod 8).

⊥ ⊥
Proof. Note that the defining sets of C(0,3,m) and C(1,2,m) with respect to α are {0} ∪ T(0,3,m) and
⊥
{0} ∪ T(1,2,m) , respectively. It then follows that C(0,3,m) is the even-weight subcode of C(0,3,m)
 28

⊥
and C(1,2,m) is the even-weight subcode of C(1,2,m) . The desired conclusion then follows from
Theorem 3.6.

When m ≡ 1 (mod 4) ≥ 5, the parameters of extended codes C(0,3,m) and C(1,2,m) are
investigated in the following theorem.

Theorem 3.8. Let m ≡ 1 (mod 4) ≥ 5 be an integer. Then the extended codes C(0,3,m) and
C(1,2,m) of C(0,3,m) and C(1,2,m) are self-dual and doubly-even. Furthermore, they have parameters
[2m , 2m−1 , d¯ ≥ 2(m−1)/2 + 4], where d¯ denotes the minimum distance of C(0,3,m) or C(1,2,m) .

Proof. It is well known that the extended codes of a pair of odd-like binary duadic codes are
self-dual if the splitting corresponding to the pair of odd-like binary duadic codes is given by
−1 [6, Theorem 6.4.12]. Note that (T(0,3,m) , T(1,2,m) , −1) is a splitting of Z2m −1 when m ≡ 1
(mod 4). Consequently, the extended codes C(0,3,m) and C(1,2,m) of C(0,3,m) and C(1,2,m) are self-
dual. It then follows from [6, Theorem 6.5.1] that the Hamming weight of each codeword in
C(0,3,m) and C(1,2,m) is divisible by 4. The remaining conclusions follow from Theorem 3.6.

Similarly, we have the following three theorems on parameters of C(0,1,m) and C(2,3,m) and
their dual and extended codes when m ≡ 3 (mod 4).

Theorem 3.9. Let m ≡ 3 (mod 4) be an integer. Then C(0,1,m) and C(2,3,m) form a pair of
odd-like duadic codes with parameters [2m − 1, 2m−1 , d], where

 2(m−1)/2 + 1 if m ≡ 3 (mod 8),
d≥
 2(m−1)/2 + 3 if m ≡ 7 (mod 8).

Proof. The proof is very similar to that of Theorem 3.6 and omitted here.

⊥ ⊥
Theorem 3.10. Let m ≡ 3 (mod 4) be an integer. Then C(0,1,m) and C(2,3,m) form a pair of
even-like duadic codes with parameters [2m − 1, 2m−1 − 1, d⊥ ], where

 2(m−1)/2 + 2 if m ≡ 3 (mod 8),
d⊥ ≥
 2(m−1)/2 + 4 if m ≡ 7 (mod 8).

Proof. The proof is very similar to that of Theorem 3.7 and omitted here.

Theorem 3.11. Let m ≡ 3 (mod 4) ≥ 7 be an integer. Then the extended codes C(0,1,m) and
C(2,3,m) of C(0,1,m) and C(2,3,m) are self-dual and doubly-even. Furthermore, they have parameters
[2m , 2m−1 , d¯ ≥ 2(m−1)/2 + 4], where d¯ denotes the minimum distance of C(0,1,m) or C(2,3,m) .
 29

Proof. The proof is very similar to that of Theorem 3.8 and omitted here.

Example 3. Let m = 5 and let α be a generator of F∗25 with α5 + α2 + 1 = 0.

⊥
• The codes C(0,3,5) and C(0,3,5) have parameters [31, 16, 7] and [31, 15, 8], respectively, where
the former code is almost optimal in the sense that the minimum distance of the optimal
binary linear code with length 31 and dimension 16 is 8, and the latter code is optimal
according to the Database [8]. In fact, C(0,3,5) has the same parameters as the best binary
cyclic code according to Table A.13 in [2].
⊥
• The codes C(1,2,5) and C(1,2,5) have parameters [31, 16, 7] and [31, 15, 8], respectively. The
⊥
comments on the parameters of C(0,3,5) and C(0,3,5) above apply here to the parameters of
⊥
C(1,2,5) and C(1,2,5) .

Example 4. Let m = 7 and let α be a generator of F∗27 with α7 + α + 1 = 0.

⊥
• The codes C(0,1,7) and C(0,1,7) have parameters [127, 64, 15] and [127, 63, 20], respectively.
⊥
• The codes C(2,3,7) and C(2,3,7) have parameters [127, 64, 15] and [127, 63, 20], respectively.
Note that C(0,1,7) and C(2,3,7) have the same parameters as the punctured binary Reed-Muller
code PRM2 (3, 7) of length 127 and order 3. The best binary duadic code known of length 127
and dimension 64 has minimum distance 19 [14].

The lower bounds on the minimum distances of C(0,1,m) , C(2,3,m) , C(0,3,m) , C(1,2,m) and their
duals developed in this paper are very close to the square-root bound. Hence, they are very
good codes in general. Example 3 shows that C(0,3,m) and C(1,2,m) could be a best cyclic code.
Example 4 shows that the minimum distance of C(0,1,7) and C(2,3,7) is less than that of the two
⊥ ⊥ ⊥
duadic codes C(0,2,7) and C(1,3,7) presented in [14], but their duals C(0,1,7) and C(2,3,7) , C(0,2,7) and
⊥
C(1,3,7) have the same parameters [127, 63, 20]. Example 4 shows that the minimum distance of
C(0,1,7) and C(2,3,7) equals that of the punctured binary Reed-Muller code PRM2 (3, 7) of length
⊥ ⊥
127 and order 3, but C(0,1,7) and C(2,3,7) are much better than PRM2 (3, 7)⊥ as

d(PRM2 (3, 7)⊥ ) = 16

and
⊥ ⊥
d(C(0,1,7) ) = d(C(2,3,7) ) = 20,

where d(C) denotes the minimum distance of C.

 30

E. Differences among several families of duadic codes

None of the families of duadic codes, C(0,1,m) , C(2,3,m) , C(0,3,m) and C(1,2,m) , is identical with
the family of binary quadratic residue codes, as 2m − 1 is composite for many odd m. The two
families of binary duadic codes studied in [14] are C(0,2,m) , C(1,3,m) and thus are not identical
with any of the families of duadic codes investigated in this paper, as their defining sets are
different. This is also justified by the facts that

d(C(0,1,7) ) = d(C(2,3,7) ) = 15

and
d(C(0,2,7) ) = d(C(1,3,7) ) = 19,

where d(C) denotes the minimum distance of C.

The family of punctured binary Reed-Muller codes PRM2 ((m − 1)/2, m) of length 2m − 1
and order (m − 1)/2 is not identical with any of the families of duadic codes investigated in
this paper, as their defining sets are different. This is also justified by the facts that
⊥ ⊥
d(C(0,1,7) ) = d(C(2,3,7) ) = 20

and
d(PRM2 (3, 7)⊥ ) = 16.

4. S UMMARY AND CONCLUDING REMARKS

In this paper, we constructed and studied the parameters of the five families of binary cyclic
codes with parameters [n, k, d] and their duals, where n = 2m −1 and (n−6)/3 ≤ k ≤ (n+6)/3.
They contain some distance-optimal codes and are very good in general, as they have a very
good lower bound on their minimum distances. The work on the two families of duadic codes
C(0,1,m) , C(2,3,m) , C(0,3,m) and C(1,2,m) complements the work in [14]. It is possible to improve the
lower bounds on the binary cyclic codes developed in this paper.
The works of [14] and this paper can be generalised and extended to obtain more families
of binary duadic codes and other binary cyclic codes. But it will be more difficult to develop a
good lower bound on their minimum distances. The generalisation goes as follows. Let r ≥ 2
be a positive integer and let n = 2m − 1 for an integer m ≥ 3. Let S be any proper subset of
Zr . Define
T[r,m,S] = {1 ≤ i ≤ n − 1 : w2 (i) mod r ∈ S}.
 31

By definition, T[r,m,S] is the union of some 2-cyclotomic cosets modulo n. Let α be a primitive
element of F2m . Let C[r,m,S] denote the binary cyclic code of length n with generator polynomial
Y
g[r,m,S] (x) = (x − αi ).
i∈T[r,m,S]

When r = 2 and |S| = 1, the codes C[r,m,S] were studied in [14]. When r = 3 and |S| = 1, the
codes C[r,m,S] were treated in this paper. When r = 4 and |S| = 2, some of the codes C[r,m,S]
were investigated in this paper, and the others were not studied in this paper as they are not
duadic codes. When r ≥ 6 is even and |S| = r/2, the code C[r,m,S] could be a duadic code for
certain odd m. Hence, we propose the following research problems.

Open Problem 1. Let m ≥ 3 be an integer and let r = 4. Let S ∈ {(0, 1), (2, 3)}, Determine
the parameters of the code C[r,m,S] for m 6≡ 3 mod 4.

Open Problem 2. Let m ≥ 3 be an integer and let r = 4. Let S ∈ {(0, 3), (1, 2)}. Determine
the parameters of the code C[r,m,S] for m 6≡ 1 mod 4.

The two problems above can be solved with similar techniques in this paper. The following
two research problems are harder.

Open Problem 3. Let m ≥ 3 be an integer and let r ≥ 5 be an integer. For any proper subset
S of Zr , determine the parameters of the code C[r,m,S] .

Open Problem 4. Let r ≥ 6 be an even integer. Find a subset S of Zr with |S| = r/2 such that
C[r,m,S] is a binary duadic code of length n = 2m − 1 for infinitely many odd m. Determine the
parameters of these duadic codes.

R EFERENCES

[1] P. Charpin, “Open problems on cyclic codes,” in Handbook of Coding Theory, vol. 1, V. S. Pless and W. C. Huffman,
Eds. Amsterdam, The Netherlands: Elsevier, 1998, pp. 963–1063.
[2] C. Ding, Codes from Difference Sets. Singapore: World Scientific, 2018.
[3] C. Ding, K. Y. Lam and C. Xing, “Enumeration and construction of all duadic codes of length pm ,” Fundamenta
Informaticae, vol. 38, no. 1, pp. 149–161, 1999.
[4] C. Ding, V. Pless, “Cyclotomy and duadic codes of prime lengths,” IEEE Trans. Inf. Theory, vol. 45, no. 2, pp. 453–466,
March 1999.
[5] C. Ding and J. Yang, “Hamming weights in irreducible cyclic codes,” Discrete Math., vol. 313, no. 4, pp. 434–446, 2013.
[6] W. C. Huffman and V. Pless, Fundamentals Error-Correcting Codes. Cambridge, U.K.: Cambridge Univ. Press, 2003.
 32

[7] B. Gong, C. Ding and C. Li, “The dual codes of several classes of BCH codes,” IEEE Trans. Inf. Theory, vol. 68, no. 2,
pp. 953–964, 2022.
[8] M. Grassl, Bounds on the minimum distance of linear codes and quantum codes, Online available at
http://www.codetables.de, accessed on 2023-1-1.
[9] J. S. Leon, J. M. Masley, and V. Pless, “Duadic codes,” IEEE Trans. Inf. Theory, vol. 30, no. 5, pp. 709–714, Sept. 1984.
[10] J. S. Leon, “A probabilistic algorithm for computing minimum weight of large error-correcting codes,” IEEE Trans Inf.
Theory, vol. 34, no. 5, pp. 1354–1359, Sept. 1988.
[11] S. Li, “The minimum distance of some narrow-sense primitive BCH codes,” SIAM J. Discrete Math., vol. 31, no. 4, pp.
2530–2569, 2017.
[12] S. Noguchi, X.-N. Lu, M. Jimbo, Y. Miao, “BCH codes with minimum distance proportional to code length, SIAM J.
Discrete Math., vol. 35, no. 1, pp. 179–193, 2021.
[13] V. Pless, J. M. Masley, and J. S. Leon, “On weights in duadic codes,” J. Comb. Theory Ser. A, vol. 44, pp. 6–21, 1987.
[14] C. Tang, C. Ding, “Binary [n, (n + 1)/2] cyclic codes with good minimum distances,” IEEE Trans. Inf. Theory, vol. 68,
no. 12, pp. 7842–7849, 2022.
[15] X. Shi, Q. Yue, Y. Wu, “The dual-containing primitive BCH codes with the maximum designed distance and their
applications to quantum codes,” Des. Codes and Cryptogr., vol. 87, pp. 2165–2183, 2019.
[16] M. Xiong, “On cyclic codes of composite length and the minimum distance,” IEEE Trans. Inf. Theory, vol. 64, no. 9, pp.
6305–6314, 2018.
[17] M. Xiong, A. Zhang, “On cyclic codes of composite length and the minimum distance II,” EEE Trans. Inf. Theory, vol.
67, no. 8, pp. 5097–5103, 2021.