The Subfield and Extended Codes of a Subclass of Optimal Three-Weight Cyclic Codes

A class of optimal three-weight [qk-1,k+1,qk-1(q-1)-1]\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$[q^k-1,k+1,q^{k-1}(q-1)-1]$$\end{document} cyclic codes over IFq\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{I\!F}}_q$$\end{document}, with k≥2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k\ge 2$$\end{document}, achieving the Griesmer bound, was presented by Heng and Yue (IEEE Trans Inf Theory 62(8):4501–4513, 2016. https://doi.org/10.1109/TIT.2016.2550029). In this paper we study some of the subfield codes of this class of optimal cyclic codes when k=2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k=2$$\end{document}. The weight distributions of the subfield codes are settled. It turns out that some of these codes are optimal and others have the best known parameters. The duals of the subfield codes are also investigated and found to be almost optimal with respect to the sphere-packing bound. In addition, the covering structure for the studied subfield codes is determined. Some of these codes are found to have the important property that any nonzero codeword is minimal, which is a desirable property that is useful in the design of a secret sharing scheme based on a linear code. Moreover, a specific example of a secret sharing scheme based on one of these subfield codes is given. Finally, a class of optimal two-weight linear codes over IFq\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{I\!F}}_q$$\end{document}, achieving the Griesmer bound, whose duals are almost optimal with respect to the sphere-packing bound is presented. Through a different approach, this class of optimal two-weight linear codes was reported very recently by Heng (IEEE Trans Inf Theory 69(2):978–994, 2023. https://doi.org/10.1109/TIT.2022.3203380). Furthermore, it is shown that these optimal codes can be used to construct strongly regular graphs.


Introduction
Let IF q be the finite field with q elements.An [n, l, d] linear code, C, over IF q is an l-dimensional subspace of IF n q with minimum Hamming distance d.It is called cyclic if (c 0 , c 1 , . . ., c n−1 ) ∈ C implies (c n−1 , c 0 , . . ., c n−2 ) ∈ C. Further, the linear code C is called optimal if there is no [n, l, d ] code with d > d, and almost optimal if there is an optimal [n, l, d + 1] code [1].
Recently, a class of optimal three-weight [q k − 1, k + 1, q k−1 (q − 1) − 1] cyclic codes over IF q achieving the Griesmer bound was presented in [2], which generalizes a result in [3] from k = 2 to arbitrary positive integer k ≥ 2. Further, the q 0 -ary subfield codes of two families of q-ary optimal linear codes were studied in [4], with q 0 being a power of a prime such that q is in turn a power of q 0 (that is, IF q 0 is a proper subfield of IF q ).Also, some basic results on subfield codes were derived and the subfield codes of ovoid codes were determined in [5].In addition, the subfield codes of several families of linear codes were obtained in [6], and the subfield codes of hyperoval and conic codes were studied in [7].The basic idea in these last four references is to consider the subfield code of an optimal, or almost optimal, linear code over IF q and expect the subfield code over IF q 0 to have also good parameters.In all cases, subfield codes with very attractive parameters were found.
Thus, the first objective of this paper is to study the q 0 -ary subfield codes for a subclass of the optimal three-weight cyclic codes reported in [2] and determine their weight distributions.It turns out that the studied subfield codes also have three nonzero weights, which is of interest as linear codes with few weights have a wide range of applications in many research fields such as authentication codes [8], secret sharing schemes [9][10][11][12][13], association schemes [14], strongly walk-regular graphs [15,16], and design of frequency hopping sequences [17].As we will see, some of the subfield codes are optimal and others have the best known parameters.The duals of the subfield codes are also investigated and found to be almost optimal with respect to the sphere-packing bound.
The second objective is to determine the covering structure for the studied subfield codes.By means of the Ashikhmin-Barg Lemma (see [18]) we show that some of these codes have the important property that all their nonzero codewords are minimal, which is a desirable property that is useful in the design of a secret sharing scheme based on a linear code.Moreover, a specific example of a secret sharing scheme based on one of these subfield codes is given.
Finally, the third objective is to present a class of optimal two-weight linear codes over IF q , achieving the Griesmer bound, whose duals are almost optimal with respect to the sphere-packing bound.This class of codes is obtained by extending some of the optimal three-weight cyclic codes reported in [2].It is important to highlight that, by means of a different approach, this class of optimal two-weight linear codes was also obtained very recently in [19,Theorem 6.3].Furthermore, it is shown that these optimal codes can be used to construct strongly regular graphs.This work is organized as follows: In Sect.2, we fix some notation and recall some definitions and some known results to be used in subsequent sections.Section 3 is devoted to presenting preliminary results.In Sect. 4 we determine the subfield codes of a subclass of already known optimal three-weight cyclic codes.In Sect.5, we investigate the covering structure for the studied subfield codes and present a specific example of a secret sharing scheme based on one of these codes.In Sect.6, a class of optimal two-weight linear codes over any finite field whose duals are almost optimal is presented.Moreover, it is shown that these codes can be used to construct strongly regular graphs.Finally, Sect.7 is devoted to conclusions.

Notation, Definitions and Known Results
Unless otherwise specified, throughout this work we will use the following: Notation.Let q 0 = p t , where t is a positive integer and p is a prime number.For an integer r > 1 we are going to fix q = q r 0 = p tr .For an integer k > 1, let IF q k be the finite extension of degree k of the finite field IF q and let γ be a primitive element of IF q k .Let F be a finite field of characteristic p and E a finite extension of F. Then we will denote by "Tr E/F " the trace mapping from E to F, while "Tr" will denote the absolute trace mapping from E to the prime field IF p .
The weight enumerator of a linear code C of length n is defined as 1 Let C be a linear code of length n over IF q .The dual code, C ⊥ , of C is the linear code defined by where •,• denotes the standard inner product in the vector space IF n q .It is known that if C is an [n, l] linear code, then C ⊥ is an [n, n − l] linear code.Moreover, the linear code C is said to be projective if the minimum Hamming distance of C ⊥ is at least 3. Now, denote by the sequence {1, A ⊥ 1 , . . ., A ⊥ n } the weight distribution of the dual code C ⊥ and fix m = n(q − 1).Then, the first four Pless power moments (see [20, pp. 259-260]) for C are: The Pless power moments relate the weight distribution of a linear code to that of its dual code.In this paper we use these identities to determine, given the weight distribution of a linear code, the minimum Hamming distance of its dual code.
When constructing an [n, l, d] code over IF q , from an economical point of view, it is desirable that its length n be minimal for given values of l, d and q.A lower bound for the length n in terms of these values is as follows (see for example [20, Theorem 2.7.4,p. 81]): where x denotes the smallest integer greater than or equal to x.
Another well-known bound for linear codes was given in [21,Theorem 6,p. 19].
where x denotes the largest integer less than or equal to x.
The sphere-packing bound is useful, for example, to find out if a code with certain parameters exists.In the present work we use it to determine the maximum value that the minimum Hamming distance of a code can take given its length and dimension.
The canonical additive character of IF q is defined as follows Let a ∈ IF q .The orthogonality relation for the canonical additive character χ of IF q is given by (see for example [22,Chapter 5]): This property plays an important role in numerous applications of finite fields.Among them, this property is useful for determining the Hamming weight of a given vector over a finite field; for example, if w(•) stands for the usual Hamming weight function and if v = (a 0 , a 1 , . . ., We now recall the class of optimal three-weight cyclic codes for which we are interested in obtaining their subfield and extended codes. Theorem 3 [2,Theorem 11] Let e 1 and e 2 be positive integers and let C (q,k,e 1 ,e 2 ) be the cyclic code of length q k − 1 over IF q given by where ) is an optimal three-weight [q k − 1, k + 1, q k−1 (q − 1) − 1] cyclic code, achieving the Griesmer bound, with weight enumerator 1 + (q − 1)(q k − 1)z q k−1 (q−1)−1 + (q k − 1)z q k−1 (q−1) + (q − 1)z q k −1 . (4) In addition, if q > 2, its dual code is a [q k − 1, q k − k − 2, 3] cyclic code.
Remark 1 In [23, Theorem 1] it was shown that the integers e 1 and e 2 in the previous theorem can be any integers.
Let C be an [n, l] linear code over IF q .The following describes a way to construct a new [n, l ] linear code, C (q 0 ) , over IF q 0 (see [5]).Let G be a generator matrix of C. Take a basis of IF q = IF q r 0 over IF q 0 and represent each entry of G as an r × 1 column vector of IF r q 0 with respect to this basis.Replace each entry of G with the corresponding r × 1 column vector of IF r q 0 .With this method, G is modified into an lr × n matrix over IF q 0 generating a new linear code, C (q 0 ) , over IF q 0 of length n, called subfield code.It is known that the subfield code C (q 0 ) is independent of both the choice of the basis of IF q over IF q 0 and the choice of the generator matrix G of C (see [5,Theorems 2.1 and 2.6]).Also, it is clear that the dimension l of C (q 0 ) satisfies l ≤ lr.

Remark 2
We recall that the subfield subcode of a linear code, C, over IF q is the subset of codewords in C whose components are all in IF q 0 (see for example [20, p. 116]).In consequence, observe that a subfield code and a subfield subcode are different codes in general.In addition, note that the subfield codes defined here are also different from the subfield codes in [24,  For what follows, we are interested in obtaining the weight distributions for the subfield codes of a subclass of the optimal three-weight cyclic codes in Theorem 3. To that end, the following is a useful result that will allow us to represent the q 0 -ary subfield code, C (q 0 ) , of a q-ary linear code, C, in terms of the trace function.
Lemma 1 [5,Theorem 2.5] Let C be an [n, l] linear code over IF q .Let G = [g i j ] 1≤i≤l,1≤ j≤n be a generator matrix of C.Then, the trace representation of the subfield code C (q 0 ) is given by
In order to obtain the weight distributions of the subfield codes of the form C (q 0 ) (q,2,e 1 ,e 2 ) , we will need the following preliminary result.
Further, if a = 0 and b = 0, Now, let ϕ be the canonical additive character of IF q 0 and suppose that (a, b) = (0, 0).By the transitivity and linearity of the trace function, we have where the last equality holds because Tr IF q 2 /IF q (bx) = bx + b q x q .Let B = IF q 2 \{ b q a }.Thus, after applying the variable substitution x → w − b q a , we obtain However, b q 2 = b and a q = a.Thus, since where χ is the canonical additive character of IF q (note that w q+1 ∈ IF q ).But, since a, y = 0, we have Therefore, finally, we obtain

The Subfield Codes of a Subclass of Optimal Cyclic Codes
By means of the following result we now determine the subfield codes, along with their weight distributions, for a subclass of the optimal three-weight cyclic codes in Theorem 3.
Similarly, the frequencies of the other weights of C (q 0 ) (q,2,e 1 ,e 2 ) can be computed and we omit the details here.Then the weight enumerator of C (q 0 ) (q,2,e 1 ,e 2 ) follows.Finally, note that A ⊥ 1 = 0, since otherwise C (q 0 ) (q,2,e 1 ,e 2 ) would be the null code {0}.Thus, a direct application of the last two identities in (1) shows that A ⊥ 2 = 0 and that the value of A ⊥ 3 is the announced one.Lastly, by the sphere-packing bound (see Theorem 2), it is not difficult to verify that for a code of length q 2r 0 − 1 and dimension q 2r 0 − 3r − 1, its minimum Hamming distance can be at most 4. Therefore, the code C (q 0 )⊥ (q,2,e 1 ,e 2 ) is almost optimal since its minimum Hamming distance is 3.

Example 1
The following are some examples of Theorem 4.
By fixing k = 2, it is important to observe that the condition on the integer e 2 is more restrictive in Theorem 4 (gcd(q 2 − 1, e 2 ) = 1) than in Theorem 3 (gcd(q + 1, e 2 ) = 1).This implies, of course, that Theorem 4 can only determine the subfield codes for a subclass of the optimal three-weight cyclic codes in Theorem 3. Specifically, this means that there are optimal three-weight cyclic codes in Theorem 3, whose subfield codes cannot be described through Theorem 4. For example, with the help of a computer, it is not difficult to verify that the subfield code C (2) (4,2,1,3) is a four-weight binary cyclic code with weight enumerator 1 + 25z 6 + 30z 8 + 3z 10 + 5z 12 (for this example note that gcd(q +1, e 2 ) = 1, but gcd(q 2 −1, e 2 ) = 1).This subfield code, like the subfield code in (b) Example 1, is optimal.However, unlike the dual of C (2) (4,2,1,1) , the dual of C (2)  (4,2,1,3) is a binary optimal cyclic code with parameters [15,9,4].This example let us know that, beyond Theorem 4, there are still other optimal three-weight cyclic codes whose subfield codes have good parameters.

The Covering Structure of the Subfield Codes
For any c = (c 0 , c 1 , . . ., c n−1 ) ∈ IF n q 0 , the support of c is defined by the set {i : 0 ≤ i ≤ n −1, c i = 0}.Furthermore, for any two vectors c, c ∈ IF n q 0 , c is said to cover c if the support of c contains that of c .A nonzero codeword is called a minimal codeword if it covers only its multiples in a linear code.The set of all minimal codewords in a linear code is called the covering structure of the code.
Determining the covering structure of a linear code is in general a difficult but at the same time interesting problem as it is closely related to the construction of secret sharing schemes (see for example [9][10][11][12][13]).In this section we determine the covering structure of the subfield codes in Theorem 4. As we will see, some of these codes have the important property that any nonzero codeword is minimal.These codes are suitable for constructing secret sharing schemes with nice access structures.Moreover, we present a specific example of a secret sharing scheme based on one of these subfield codes.
There are several ways to construct secret sharing schemes by using linear codes.One of them was proposed by Massey in [11,12] and is presented below (see [10,13]).
Let C be an [n, l] linear code over IF q 0 .In the secret sharing scheme based on a linear code C, the secret s is an element of IF q 0 , which is called the secret space.There is a dealer P 0 and n −1 parties P 1 , P 2 , . . ., P n−1 involved in the secret sharing scheme, the dealer being a trusted person.Let G ⊥ = (g ⊥ 0 , g ⊥ 1 , . . ., g ⊥ n−1 ) be a generator matrix of the dual code, C ⊥ , of C such that g ⊥ i is the i-th column vector of G ⊥ and g ⊥ i = 0 for 0 ≤ i ≤ n − 1.Then, the secret sharing scheme based on C is described as follows: Step 1) In order to compute the shares with respect to a secret s, the dealer P 0 chooses randomly a vector u = (u 0 , u 1 , . . ., u n−l−1 ) ∈ IF n−l q 0 such that s = ug ⊥ 0 .There are altogether q n−l−1 0 such vectors u ∈ IF n−l q 0 .Step 2) The dealer P 0 treats u as an information vector and computes the corresponding codeword t = uG ⊥ = (t 0 , t 1 , . . ., t n−1 ) in C ⊥ .Then he sends t i to party P i as the share for every i (1 ≤ i ≤ n − 1).
Step 3) The secret s is recovered as follows: since t 0 = ug ⊥ 0 = s, a set of shares {t i 1 , t i 2 , . . ., t i m } can determine the secret s iff g ⊥ 0 is a linear combination of Clearly, if a group of participants D can recover the secret by combining their shares, then any group of participants containing D can also recover the secret.The set {i 1 , i 2 , . . ., i m } is said to be a minimal access set if it can recover the secret s but none of its proper subsets can do so.The access structure of the secret sharing scheme refers to the set of all minimal access sets.For a linear code C, the following lemma from [11] presents a one-to-one correspondence between the set of minimal access sets of the secret sharing scheme based on C and the set of minimal codewords in C whose first coordinate is 1.

Lemma 3 Let C be an [n, l] linear code over
If c is a nonzero codeword whose first coordinate is 1 and the support of the codeword c is we call the set {i 1 , i 2 , . . ., i m } the access support of the codeword c.
From the discussion above, determining the access structure of the secret sharing scheme based on a linear code C is equivalent to determining the set of access supports of the minimal codewords in C whose first coordinate is 1.Thus, in the following we determine the covering structure of the subfield codes in Theorem 4. To that end, the next results found in [18] will be useful.

Lemma 4
Let C be a linear code over IF q 0 with minimum Hamming distance d.Then, every codeword whose weight is less than or equal to dq 0 −q 0 +1 q 0 −1 must be a minimal codeword.
The following lemma states that if the weights of a linear code are close enough to each other, then all nonzero codewords of the code are minimal.
Lemma 5 (Ashikhmin-Barg Lemma) Let C be an [n, l] linear code over IF q 0 , and let w min and w max be the minimum and maximum nonzero weights of C, respectively.If then all nonzero codewords of C are minimal.
We remark that the condition in the previous lemma is only a sufficient condition.There are codes such that all their nonzero codewords are minimal without satisfying this condition (see for example [13]).Now, we are able to give the covering structure of the subfield codes in Theorem 4: Theorem 5 Assume the same notation as in Theorem 4. Then the covering structure of a subfield code of the form C (q 0 ) (q,2,e 1 ,e 2 ) is as follows: (q,2,e 1 ,e 2 ) belongs to Part (A) of Theorem 4, then all its nonzero codewords with weight q 2r 0 − 1 are not minimal, while the other nonzero codewords are minimal.(b) If C (q 0 ) (q,2,e 1 ,e 2 ) belongs to Part (B) of Theorem 4, then all its nonzero codewords are minimal.
Linear codes whose nonzero codewords are all minimal are suitable for constructing secret sharing schemes with nice access structures as described in the following: Proposition 2] Let C be an [n, l] linear code over IF q 0 and let G = (g 0 , g 1 , . . ., g n−1 ) be a generator matrix of C such that g i is the i-th column vector of G and g i = 0 for 0 ≤ i ≤ n − 1.If each nonzero codeword of C is minimal, then the access structure of the secret sharing scheme based on C is composed of q l−1 0 minimal access sets, which is equal to the set of access supports of the nonzero codewords in C with first coordinate 1.In addition, we have the following: (a) If g i is a scalar multiple of g 0 , 1 ≤ i ≤ n − 1, then participant P i must be in every minimal access set.Such a participant is called a dictatorial participant.(b) If g i is not a scalar multiple of g 0 , 1 ≤ i ≤ n − 1, then participant P i must be in (q 0 − 1)q l−2 0 out of q l−1 0 minimal access sets.
We end this section by presenting a specific example of a secret sharing scheme based on one of the subfield codes in Theorem 4.

A Class of Optimal Two-Weight Linear Codes
Let C be an [n, l, d] linear code over IF q .The extended code, C, of C is the linear code defined by (see for example [20, p. 14]) Through the following result we present a class of optimal two-weight linear codes over any finite field whose duals are almost optimal.Theorem 6 Let C (q,k,e 1 ,e 2 ) be the extended code of the cyclic code C (q,k,e 1 ,e 2 ) in Theorem 3. If e 1 = 0 (see Remark 1), then the extended code C (q,k,0,e 2 ) is an optimal two-weight [q k , k + 1, q k−1 (q − 1)] linear code over IF q , achieving the Griesmer bound, with weight enumerator 1 + q(q k − 1)z q k−1 (q−1) + (q − 1)z q k .In addition, A ⊥ 1 = A ⊥ 2 = 0, and Therefore, if q > 2, then C (q,k,0,e 2 ) is projective and its dual code, C ⊥ (q,k,0,e 2 ) , is a [q k , q k − k − 1, 3] linear code which is almost optimal with respect to the spherepacking bound.
Proof Recall that each codeword of the cyclic code C (q,k,0,e 2 ) is of the form c(a, b) = a + Tr IF q k /IF q bx e 2 x∈IF * q k , with a ∈ IF q and b ∈ IF q k .
For each codeword c(a, b) in C (q,k,0,e 2 ) , let c(a, b) denote the corresponding extended codeword in C (q,k,0,e 2 ) .
In addition, a direct application of the last three identities in (1) shows that A ⊥ 1 = A ⊥ 2 = 0 and that the value of A ⊥ 3 is the announced one.Clearly, if q > 2, then A ⊥ 3 = 0 and C (q,k,0,e 2 ) is a projective code.Finally, by the sphere-packing bound, it is not difficult to verify that for a code of length q k and dimension q k − k − 1, its minimum Hamming distance can be at most 4. Therefore, the code C ⊥ (q,k,0,e 2 ) is almost optimal.
Remark 5 A class of optimal projective two-weight linear codes with the same parameters and weight distribution as the codes in Theorem 6 was recently presented in [19,Theorem 6.3].However, unlike what was done here, this class of codes is obtained by considering a particular type of almost difference sets.Moreover, at the end of [19, Section VI] the author correctly asserts that the extended code of any cyclic code in Theorem 3 has parameters [q k , k + 1, q k−1 (q − 1) − 1] and three nonzero weights.These three-weight codes are not optimal.However, by allowing e 1 = 0 in Theorem 3, Theorem 6 shows us that the extended code obtained is a two-weight linear code which results to be optimal.

Example 3
The following are some examples of the previous theorem.
It is well known that projective two-weight linear codes are closely related to finite projective spaces and strongly regular graphs.What is remarkable is that results from one area can immediately be translated into the other two (see [16]).In the following we use the projective two-weight linear codes in Theorem 6 in order to determine the strongly regular graphs associated to them.
A graph is a set V of vertices provided with a symmetric relation ∼ on V called adjacency, such that no v ∈ V is adjacent to itself.A pair of adjacent vertices v 1 , v 2 ∈ V form an edge and, under these conditions, v 1 is said to be a neighbor of v 2 and vice versa.
Let N , K , λ and μ be integers.A graph with N vertices is said to be strongly regular with parameters (N , K , λ, μ) if it is regular of degree K (that is, each vertex has exactly K neighbors), and any two distinct vertices have λ common neighbors if they are adjacent and μ common neighbors if they are nonadjacent.
Calderbank and Kantor [16] proposed a way to construct strongly regular graphs using projective two-weight linear codes.Let w 1 and w 2 be the weights of a q-ary projective two-weight linear code C of length n and dimension l with generator matrix G.To C we associate a graph as follows.Take as vertices the elements of the vector space IF l q , where two different vertices v 1 and v 2 are adjacent iff v 1 − v 2 is a multiple of a column in G.The graph obtained in this way is strongly regular [16, Theorems 3.1 and 3.2] with the following parameters [16,Corollary 3.7]: As a direct consequence of the above, we have the following: Theorem 7 Assume the same notation as in Theorem 6.If q > 2, then the extended code C (q,k,0,e 2 ) generates a strongly regular graph with parameters (q k+1 , q k (q − 1), q k (q − 2), q k (q − 1)).
We end this section by presenting an example of the construction of a strongly regular graph from one of the extended codes in Theorem 6.

Conclusions
In this paper we studied the q 0 -ary subfield codes of a subclass of optimal three-weight cyclic codes of length q 2 − 1 and dimension 3 that belongs to the class of codes in Theorem 3. We proved that some of these subfield codes are optimal three-weight cyclic codes of length q 2 − 1 and dimension 2r + 1 (where q = q r 0 ) that belong, again, to the class of optimal three-weight cyclic codes in Theorem 3 (Part (A) of Theorem 4).For the other subfield codes studied here we showed that they are three-weight cyclic codes of length q 2 − 1 whose dimension is now 3r (Part (B) of Theorem 4).For the latter subfield codes, we also determined the minimum Hamming distance for their duals, and with this, we concluded that these duals are almost optimal with respect to the sphere-packing bound.Furthermore, it was shown that some subfield codes in Part (B) of Theorem 4 are optimal and others have the best known parameters according to the code tables at [25] (Example 1 and Remark 4).However, as pointed out at the end of Sect.4, there is evidence of the existence of other subfield codes with good parameters.Therefore, as further work, it would be interesting to study those other subfield codes.
In adittion, as an application of linear codes with few weights, the covering structure of the subfield codes in Theorem 4 was determined (Theorem 5) and used to present a specific example of a secret sharing scheme based on one of these subfield codes at the end of Sect. 5 (Example 2).
Finally, by extending some of the optimal three-weight cyclic codes in Theorem 3, a class of optimal two-weight linear codes over IF q , achieving the Griesmer bound, whose duals are almost optimal with respect to the sphere-packing bound was presented (Theorem 6).Through the analysis of several examples it is suggested that such duals are optimal (Example 3 and Remark 6).We also used the extended codes in Theorem 6 to construct strongly regular graphs (Theorem 7 and Example 4).It is important to note that the parameters and weight distribution of the codes in Theorem 6 are the same as those of a class of codes in Example SU1 in [16] (therein l = k + 1 and t = k).Also, as pointed out in Remark 5, they are the same for the class of codes in [19,Theorem 6.3].Thus, as future work, it would be interesting to show if these codes are equivalent to each other.