In the literature, no encoding method has been given for symmetric, alternating and Hermitian d-codes. This section is dedicated to the encoding of these three types of restricted d-codes. As a matter of fact, the encoding of an optimal d-code \(\mathcal {C}\) is mainly concerned with setting up a one-to-one correspondence between a message space of size \(\#\mathcal {C}\) and the code \(\mathcal {C}\) in an efficient way, which ideally also allows for an efficient decoding algorithm.
3.1 Encoding of symmetric d-codes
We start with the encoding of the optimal symmetric d-codes of size qn(n−d+ 2)/2 in Theorem 4, where n − d is even. The family of codes is linear over \(\mathbb {F}_{q}\) and the message space is naturally a vector space over \(\mathbb {F}_{q}\) with dimension n(n − d + 2)/2. But we can represent each message in the form of a k-dimensional vector over \(\mathbb {F}_{q^{n}}\) where k = (n − d + 2)/2 and the set of all the message vectors are closed under \(\mathbb {F}_{q}\)-linear operations.
In order to present a polynomial-time decoding algorithm for the optimal symmetric
d-codes in Theorem 4, we shall express their codewords as evaluations of certain polynomials at linearly independent points over
\({\mathbb {F}}_{q}\). For this reason, we need to employ a pair of dual bases in
\({\mathbb {F}}_{q^{n}}\) over
\({\mathbb {F}}_{q}\). Recall that given an ordered
\({\mathbb {F}}_{q}\)-basis (
α1,…,
αn) of
\({\mathbb {F}}_{q^{n}}\), its dual basis is defined as the ordered
\({\mathbb {F}}_{q}\)-basis (
β1,…,
βn) of
\({\mathbb {F}}_{q^{n}}\) such that
$$\text{Tr}_{q^{n}/q}(\alpha_{i}\beta_{j}) = \delta_{ij} \text{ for } i = 1, 2, \dots, n,$$
where
δij denotes the Kronecker delta function. Note that a dual basis always exists for a given order basis (
α1,…,
αn) of
\({\mathbb {F}}_{q^{n}}\) [
15, Definition 2.30].
Let (
α1,…,
αn), (
β1,…,
βn) be a pair of dual bases of
\(\mathbb {F}_{q^{n}}\) over
\({\mathbb {F}_{q}}\). We will write
\(\text {Tr}_{q^{n}/q}(x)\) as Tr(
x) for simplicity when the context is clear. Let
L(
x) be a linearized polynomial as in Theorem 4. For the symmetric form we have
$$S(x,y)=\text{Tr}(x L(y)).$$
Now, we denote the associated matrix of
S with respect to the ordered
\(\mathbb {F}_{q}\)-basis (
α1,…,
αn) by
\(\mathcal {S}\), of which the (
i,
j)-th entry
\(\mathcal {S}(i,j)\) is given by
$$\mathcal{S}(i,j)=S(\alpha_{i}, \alpha_{j})=\text{Tr}(\alpha_{j} L(\alpha_{i})) .$$
Furthermore, the codewords of the additive
d-code in Theorem 4 can be expressed in the symmetric matrix form as follows: let
\(x, y \in {\mathbb {F}}_{q^{n}}\), then
\(x=\sum \limits _{i=1}^{n} x_{i}\alpha _{i}\) and
\(y=\sum \limits _{j=1}^{n} y_{j}\alpha _{j}\) for some
\(x_{i},y_{j} \in {\mathbb {F}}_{q}\) and
$$\begin{array}{@{}rcl@{}} S(x,y)&=&\text{Tr}\left( \left( \sum\limits_{j}y_{j}\alpha_{j}\right)\sum\limits_{i}x_{i}L(\alpha_{i})\right) =\text{Tr}\left( \sum\limits_{i,j}x_{i}y_{j}\alpha_{j}L(\alpha_{i})\right)\\ &=&\sum\limits_{i,j}x_{i}y_{j}\text{Tr}\left( \alpha_{j}L(\alpha_{i})\right) = \sum\limits_{i,j}x_{i}\mathcal{S}(i,j)y_{j} =(x_{1},\ldots,x_{n})\cdot \mathcal{S}\cdot \left( \begin{array}{c} y_{1} \\ y_{2}\\ \vdots\\ y_{n} \end{array}\right), \end{array}$$
where \(\mathcal {S}(i,j)\) is the (i,j)-th entry in \(\mathcal {S}\).
In the following we show that the evaluation of the corresponding linearized polynomial at linearly independent elements α1,…,αn is a proper encoding method.
Define an
n-dimensional vector over
\(\mathbb {F}_{q}\) as
$$s=(s_{1},\ldots,s_{n})=(\beta_{1},\ldots,\beta_{n})\cdot \mathcal{S}^{T}.$$
Since the
i-th row of
\(\mathcal {S}\) is given by (Tr(
α1L(
αi)),…,Tr(
αnL(
αi))) and since each
L(
αi) can be written as
\({\sum }_{t} c_{t} \beta _{t}\) for some
\(c_{t} \in {\mathbb {F}}_{q}\), we can write
si as
$$\begin{array}{@{}rcl@{}} s_{i} &=&\sum\limits_{j}\beta_{j} S(i,j) =\sum\limits_{j}\beta_{j} \text{Tr}(\alpha_{j}L(\alpha_{i})) \\ & =&\sum\limits_{j}\beta_{j}\text{Tr}\left( \alpha_{j}\sum\limits_{t}c_{t}\beta_{t}\right) =\sum\limits_{j}\beta_{j}\sum\limits_{t}c_{t}\text{Tr}(\alpha_{j}\beta_{t}) \\& =&\sum\limits_{j}\beta_{j}c_{j}=L(\alpha_{i}) \end{array}$$
since Tr(x) is linear over \(\mathbb {F}_{q}\) and (β1,…,βn) is the dual basis of (α1,…,αn). From the equality si = L(αi), we see that the encoding of symmetric d-codes given by Tr(yL(x)), as in Theorems 4 and 5, can be seen as the evaluation of L(x) at the basis (α1,…,αn) of \(\mathbb {F}_{q^{n}}\).
With the above preparation, we are now ready to look at the encoding of the optimal symmetric d-codes in Theorem 4 more explicitly.
Let (
ω0,…,
ωn− 1) be a basis of
\(\mathbb {F}_{q^{n}}\) over
\(\mathbb {F}_{q}\). For optimal symmetric
d-codes in Theorem 4, the linearized polynomial can be expressed as
$$L(x) = b_{0}x+\sum\limits_{j=1}^{k-1} \left( b_{j}x^{[j]}+(b_{j}x)^{[n-j]} \right),$$
where
k = (
n −
d + 2)/2. Then the encoding of a message
\(f=(f_{0},\ldots , f_{k-1}) \in {\mathbb {F}}_{q^{n}}^{k}\) for the symmetric codes in Theorem 4 can be expressed as the evaluation of the following linearized polynomial at points
ω0,…,
ωn− 1:
$$L(x) = f_{0}x+\left( \sum\limits_{j=1}^{k-1} f_{j} x^{[j]} + (f_{j}x)^{{[n-j]}}\right) = \sum\limits_{i=0}^{n-1}\tilde{f}_{i}x^{[i]},$$
where
$$\begin{array}{@{}rcl@{}} \tilde{f}&=&\left( \tilde{f}_{0}, \dots,\tilde{f}_{k-1},0,\ldots,0 ,\tilde{f}_{n-k+1},\ldots,\tilde{f}_{n-1}\right) \\&=& \left( f_{0}, \dots, f_{k-1}, 0, \dots, 0, f^{[n-k+1]}_{k-1}, \dots, f^{[n-1]}_{1}\right). \end{array}$$
(6)
Let
\(N= \left (\begin {array}{c} \omega _{i}^{[j]} \end {array}\right )_{n\times n}\) be the
n ×
n Moore matrix generated by
ωi’s. So the encoding of optimal symmetric and optimal alternating
d-codes can be expressed as
$$(f_{0},\ldots,f_{k-1})\mapsto (L(\omega_{0}),\ldots,L(\omega_{n-1}))=\tilde{f}\cdot N^{T},$$
(7)
where
\(\tilde {f}=\left (\tilde {f}_{0}, \ldots , \tilde {f}_{n-1}\right )\) and
NT is the transpose of the matrix
N. Note that the first
k and the last
k − 1 elements of
\(\tilde {f}\) are nonzero. This means at most
n −
d + 1 columns of the matrix
NT are involved in the encoding process.
3.2 Encoding of alternating d-codes
The encoding of alternating d-codes in Theorem 5 can be done similarly since the codewords in A(x,y) has the same form Tr(yL(x)) as in Theorem 4.
For alternating
d-codes in Theorem 5, the linearized polynomial can be expressed as
$$L(x) = \sum\limits_{j=e}^{\frac{n-1}{2}} \left( b_{j}x^{[j]}-(b_{j}x)^{[n-j]} \right).$$
Note that in Theorem 5, the parameters
n is odd and
d = 2
e. The optimal alternating codes are
\(\mathbb {F}_{q}\)-linear with dimension
n(
n −
d + 1)/2. For simplicity, we again consider the message vectors in the form of vectors over
\(\mathbb {F}_{q^{n}}\).
Let (
ω0,…,
ωn− 1) be a basis of
\(\mathbb {F}_{q^{n}}\) over
\(\mathbb {F}_{q}\). The encoding of a message
\(f=(f_{0},\ldots , f_{k-1}) \in {\mathbb {F}}_{q^{n}}^{k}\) can be expressed as the evaluation of the following linearized polynomial at points
ω0,…,
ωn− 1:
$$L(x)=\left( \sum\limits_{j=e}^{\frac{n-1}{2}} f_{j-e} x^{[j]} - (f_{j-e}x)^{{[n-j]}}\right)=\sum\limits_{i=0}^{n-1}\tilde{f}_{i}x^{[i]},$$
where
$$\begin{array}{@{}rcl@{}} \tilde{f}&=&\left( 0, \dots,0,\tilde{f}_{e},\ldots,\tilde{f}_{\frac{n-1}{2}},\tilde{f}_{\frac{n+1}{2}},{\ldots} ,\tilde{f}_{n-e},0,\ldots,0\right) \\&=& \left( 0,\dots,0, f_{0}, \dots, f_{k-1}, -f^{[\frac{n+1}{2}]}_{k-1}, \dots, -f^{[n-e]}_{0},0,\ldots,0\right). \end{array}$$
(8)
Similarly, the encoding of optimal alternating
d-code can be expressed as
$$(f_{0},\ldots,f_{k-1})\mapsto (L(\omega_{0}),\ldots,L(\omega_{n-1}))=\tilde{f}\cdot N^{T},$$
(9)
where
\(\tilde {f}=\left (\tilde {f}_{0}, \ldots , \tilde {f}_{n-1}\right )\) and
NT is the transpose of the matrix
N. As shown in (
8), at most
n −
d + 1 columns of the matrix
N are involved in computation.
3.3 Encoding of Hermitian d-codes
This section is dedicated to the encoding of the optimal Hermitian d-codes of size qn(n−d+ 1) explained in Theorems 7 and 8. Given positive integers d,n with 1 ≤ d ≤ n, for encoding of optimal Hermitian d-codes we are going to set up a one-to-one correspondence between a message space of size qn(n−d+ 1), and a Hermitian optimal d-code, which later permits us to decode efficiently. Therefore, for a message space of size qn(n−d+ 1), we may assume its elements as vectors over \({\mathbb {F}}_{q^{n}}\) of dimension k = n − d + 1.
For the optimal Hermitian
d-codes in Theorems 7 and 8, we shall express their codewords as evaluations of certain polynomials at linearly independent points over
\({\mathbb {F}}_{q^{2}}\). For this reason, we need to introduce the Hermitian variant of a basis in
\({\mathbb {F}}_{q^{2n}}\) over
\({\mathbb {F}}_{q^{2}}\). Given an ordered
\({\mathbb {F}}_{q^{2}}\)-basis (
α1,…,
αn) of
\({\mathbb {F}}_{q^{2n}}\), its
Hermitian dual basis is defined as the ordered
\({\mathbb {F}}_{q^{2}}\)-basis (
β1,…,
βn) of
\({\mathbb {F}}_{q^{2n}}\) such that
$$\text{Tr}_{q^{2n}/q^{2}}\left( {\alpha_{i}^{q}}\beta_{j}\right) = \delta_{ij} \text{ for } i = 1, 2, \dots, n,$$
where
\(\text {Tr}_{q^{2n}/q^{2}}\) is the relative trace function from
\(\mathbb {F}_{q^{2n}}\) to
\(\mathbb {F}_{q^{2}}\), namely,
\(\text {Tr}_{q^{2n}/q^{2}}(x)=\sum \limits _{i=0}^{n-1}x^{q^{2i}}\) and
δij denotes the Kronecker delta function. Note that such a Hermitian dual basis always exists for a given order basis (
α1,…,
αn). Indeed, since there exist a dual basis (
γ1,…,
γn) for (
α1,…,
αn) satisfying
\(\text {Tr}_{q^{2n}/q^{2}}(\alpha _{i}\gamma _{j})=\delta _{ij}\), one can simply takes
\(\beta _{j} = \gamma _{j}^{q^{2n-1}}\) for
\(j=1,2,\dots , n\) and then the above Hermitian dual property follows. We shall also write
\(\text {Tr}_{q^{2n}/q^{2}}()\) as Tr() for simplicity whenever there is no ambiguity.
Let (α1,…,αn) be an \(\mathbb {F}_{q^{2}}\)-basis of \(\mathbb {F}_{q^{2n}}\) and (β1,…,βn) be its Hermitian dual as described above. Let \(x,y \in {\mathbb {F}}_{q^{2n}}\), then \(x=\sum \limits _{i=1}^{n} x_{i} \alpha _{i}\) and \(y=\sum \limits _{i=1}^{n} y_{i} \beta _{i}\), for some \(x_{i},y_{i} \in {\mathbb {F}}_{q^{2}}\). It is clear that \(\text {Tr}(x^{q}y)=\sum \limits _{i,j=1}^{n}{x_{i}^{q}}y_{j}\text {Tr}({\alpha _{i}^{q}}\beta _{j})=\sum \limits _{i=1}^{n}{x_{i}^{q}}y_{i}=\langle ({x_{1}^{q}},\ldots ,{x_{n}^{q}}),(y_{1},\ldots ,y_{n})\rangle\).
Note that the Hermitian forms in Theorems 7 and 8 are of the form
H(
x,
y) = Tr(
xqL(
y)). Now, we denote the associated matrix of
H with respect to the ordered
\(\mathbb {F}_{q^{2}}\)-basis (
α1,…,
αn) by
\({\mathcal{H}}\), of which the (
i,
j)-th entry
\({\mathcal{H}}(i,j)\) is given by
$$\mathcal{H}(i,j)=H(\alpha_{i}, \alpha_{j})=\text{Tr}\left( {\alpha_{j}^{q}} L(\alpha_{i})\right) .$$
Furthermore, the codewords of the additive
d-code in Theorem 8 can be expressed in the Hermitian matrix form as follows
$$\begin{array}{@{}rcl@{}} H(x,y)&=&\text{Tr}\left( \left( \sum\limits_{j}y_{j}\alpha_{j}\right)^{q}\sum\limits_{i}x_{i}L(\alpha_{i})\right) =\text{Tr}\left( \sum\limits_{i,j}x_{i}{y_{j}^{q}}{\alpha_{j}^{q}}L(\alpha_{i})\right)\\ &=&\sum\limits_{i,j}x_{i}{y_{j}^{q}}\text{Tr}\left( {\alpha_{j}^{q}}L(\alpha_{i})\right) = \sum\limits_{i,j}x_{i}\mathcal{H}(i,j){y_{j}^{q}} =(x_{1},\ldots,x_{n})\cdot \mathcal{H}\cdot \left( \begin{array}{c} {y_{1}^{q}} \\ {y_{2}^{q}}\\ \vdots\\ {y_{n}^{q}} \end{array}\right), \end{array}$$
where
\({\mathcal{H}}(i,j)\) is an element in
\({\mathcal{H}}\). In the following we show that the evaluation of the corresponding linearized polynomial at linearly independent elements
α1,…,
αn is a proper encoding method. Define an
n-dimensional vector over
\({\mathbb {F}}_{q^{2}}\) as
$$h=(h_{1},\ldots,h_{n})=(\beta_{1},\dots,\beta_{n})\cdot \mathcal{H}^{T}.$$
Since the
i-th row of
\({\mathcal{H}}\) is given by
\((\text {Tr}({\alpha _{1}^{q}}L(\alpha _{i})), \ldots , \text {Tr}({\alpha _{n}^{q}}L(\alpha _{i})))\) and since each
L(
αi) can be written as
\({\sum }_{t} c_{t}\beta _{t}\) for some
\(c_{t} \in {\mathbb {F}}_{q^{2}}\), we can write
hi as
$$\begin{array}{@{}rcl@{}} h_{i} &=&\sum\limits_{j}\beta_{j} H(i,j) =\sum\limits_{j}\beta_{j} \text{Tr}\left( {\alpha_{j}^{q}}L(\alpha_{i})\right) \\ & =&\sum\limits_{j}\beta_{j}\text{Tr}\left( {\alpha_{j}^{q}}\sum\limits_{t}c_{t}\beta_{t}\right) =\sum\limits_{j}\beta_{j}\sum\limits_{t}c_{t}\text{Tr}\left( {\alpha_{j}^{q}}\beta_{t}\right) \\& = & \sum\limits_{t}\beta_{t}c_{t}=L(\alpha_{i}), \end{array}$$
where the fourth and fifth equality signs hold because Tr(x) is linear over \(\mathbb {F}_{q^{2}}\) and (β1,…,βn) is the Hermitian dual basis of (α1,…,αn). From the equality hi = L(αi), we see that the encoding of Hermitian d-codes given by Tr(yqL(x)), as in Theorems 7 and 8, can be seen as the evaluation of L(x) at the basis (α1,…,αn) of \(\mathbb {F}_{q^{2n}}\).
With the above preparation, we are now ready to look at the encoding of the Hermitian d-codes in Theorems 7 and 8 more explicitly.
Let
\(\kappa =\lceil \frac {n-d}{2}\rceil\) and
H be the Hermitian form given in Theorem 7. The linearized polynomial in (
4) can be written as
$$L(x) = \sum\limits_{j=1}^{\kappa} \left( (b_{j}x)^{[{\kern-1.2pt}[ n+1-j]{\kern-1.2pt}]}+{b_{j}^{q}} x^{[{\kern-1.2pt}[ j]{\kern-1.2pt}] } \right),$$
and assuming
\(m=\frac {n+1}{2}\), similarly one can write the linearized polynomial in (
5) as
$$L(x) = (b_{0}x)^{[{\kern-1.2pt}[ m]{\kern-1.2pt}] } + \sum\limits_{j=1}^{\kappa} \left( (b_{j}x)^{[{\kern-1.2pt}[ m+j]{\kern-1.2pt}] }+{b_{j}^{q}} x^{[{\kern-1.2pt}[ m-j{]{\kern-1.2pt}]}} \right).$$
Let {1,η} be an \(\mathbb {F}_{q^{n}}\)-basis of \(\mathbb {F}_{q^{2n}}\). Let α0,α1,…,αn− 1 be a basis of \({\mathbb {F}}_{q^{2n}}\) over \(\mathbb {F}_{q^{2}}\). Raising all the basis elements αi to the q2-th power will still give a linearly independent set of elements in \(\mathbb {F}_{q^{2n}}\). We use \(\alpha _{0}^{q^{2}},\alpha _{1}^{q^{2}}, \ldots , \alpha _{n-1}^{q^{2}}\) as the evaluation points for optimal Hermitian d-codes in Theorem 7. The reason for this is to keep the consistent form \(L(x)=l_{0}x^{{[{\kern -1.2pt}[} 0 {]{\kern -1.2pt}]}}+l_{1}x^{{[{\kern -1.2pt}[} 1 {]{\kern -1.2pt}]}}+\cdots +l_{n-1}x^{{[{\kern -1.2pt}[} n-1 {]{\kern -1.2pt}]}}\) for the linearized polynomial representation (employing α0,…,αn− 1 as the evaluation points for this codes will obligate us to use the linearized polynomial of the form \(L(x)=l_{0}x^{{[{\kern -1.2pt}[} 1 {]{\kern -1.2pt}]}}+l_{1}x^{{[{\kern -1.2pt}[} 2 {]{\kern -1.2pt}]}}+\cdots +l_{n-1}x^{{[{\kern -1.2pt}[} n {]{\kern -1.2pt}]}}\)).
The encoding of a message
\(f=(f_{0},\ldots , f_{k-1}) \in \mathbb {F}_{q^{n}}^{k}\) can be expressed as the evaluation of the following linearized polynomial at points
\(\alpha _{0}^{q^{2}},\alpha _{1}^{q^{2}}, \ldots , \alpha _{n-1}^{q^{2}}\):
$$L(x) = \left( \sum\limits_{j=0}^{\kappa-1} (f_{j}+\eta f_{\kappa+j})^{q}x^{[{\kern-1.2pt}[ n-1-j ]{\kern-1.2pt}]} + (f_{j}+\eta f_{\kappa+j}x)^{[{\kern-1.2pt}[ j]{\kern-1.2pt}] }\right) =\sum\limits_{i=0}^{n-1}\tilde{f}_{i}x^{[{\kern-1.2pt}[ i ]{\kern-1.2pt}]},$$
(10)
where
$$\begin{array}{@{}rcl@{}} \tilde{f}&=&\left( \tilde{f}_{0}, \dots,\tilde{f}_{\kappa-1},0,\ldots,0,\tilde{f}_{n-\kappa},\ldots, \tilde{f}_{n-1}\right) = ((f_{0}+\eta f_{\kappa})^{[{\kern-1.2pt}[ 0]{\kern-1.2pt}]},\dots,\\&&(f_{\kappa-1}+ \eta f_{2\kappa-1})^{[{\kern-1.2pt}[ \kappa-1]{\kern-1.2pt}]},0, \dots, 0, (f_{\kappa-1}+ \eta f_{2\kappa-1})^{q},\ldots,(f_{0}+ \eta f_{\kappa})^{q}), \end{array}$$
(11)
and
k = 2
κ. For the optimal Hermitian
d-code in Theorem 8 and the evaluation points
α0,
α1,…,
αn− 1, the encoding of a message
\(f=(f_{0},\ldots , f_{k-1}) \in {\mathbb {F}}_{q^{n}}^{k}\) can be expressed as the evaluation of the following linearized polynomial at points
α0,
α1,…,
αn− 1:
$$\begin{array}{@{}rcl@{}} L(x)&=& (f_{0}x)^{[{\kern-1.2pt}[ m ]{\kern-1.2pt}]} + \left( \sum\limits_{j=1}^{\kappa} \left( f_{j}+\eta f_{\kappa+j}\right)^{q}x^{[{\kern-1.2pt}[ m-j]{\kern-1.2pt}]} +\left( (f_{j}+ \eta f_{\kappa+j})x\right)^{{[{\kern-1.2pt}[} m+j{]{\kern-1.2pt}]}}\right) \\&=&\sum\limits_{i=0}^{n-1}\tilde{f}_{i}x^{[{\kern-1.2pt}[ i]{\kern-1.2pt}]}, \end{array}$$
(12)
where
$$\begin{array}{@{}rcl@{}} \tilde{f}&=&\left( 0, \dots,0,\tilde{f}_{m-\kappa},\ldots,\tilde{f}_{m-1},\tilde{f}_{m},\tilde{f}_{m+1},\ldots,\tilde{f}_{m+\kappa},0,\ldots, 0\right)\\&=& (0,\dots,0,(f_{\kappa}+\eta f_{2\kappa})^{q},\ldots,(f_{1}+\eta f_{\kappa+1})^{q},f_{0}^{[{\kern-1.2pt}[ m ]{\kern-1.2pt}]},\\&&(f_{1}+\eta f_{\kappa+1})^{{[{\kern-1.2pt}[} m+1]{\kern-1.2pt}]},\ldots,(f_{\kappa}+\eta f_{2\kappa})^{{[{\kern-1.2pt}[} m+\kappa]{\kern-1.2pt}]},0,\ldots,0 ), \end{array}$$
(13)
and
k = 2
κ + 1. The first
\(\frac {d+1}{2}\) and the last
\(\frac {d-3}{2}\) coefficients of
\(\tilde {f}\) are zero.
Let
\(M_{l}= \left (\begin {array}{c} \alpha _{i}^{[{\kern -1.2pt}[ j+l]{\kern -1.2pt}] } \end {array}\right )_{n\times n}\) be the
n ×
n Moore matrix generated by
\(\alpha _{0}^{q^{2l}},\alpha _{1}^{q^{2l}}, \ldots , \alpha _{n-1}^{q^{2l}}\) where
l ∈{0,1}. We take
l = 1 when we consider
\(\alpha _{0}^{q^{2}},\alpha _{1}^{q^{2}}, \ldots , \alpha _{n-1}^{q^{2}}\) as the evaluation points which is used in (
10) and
l = 0 when
α0,
α1,…,
αn− 1 are the evaluation points in (
12).
So the encoding of the optimal Hermitian rank metric codes can be expressed as
$$(f_{0},\ldots,f_{k-1})\mapsto \left( L\left( \alpha_{0}^{q^{2l}}\right),\ldots,L\left( \alpha_{n-1}^{q^{2l}}\right)\right)=\tilde{f}\cdot {M_{l}^{T}},$$
(14)
where
\(\tilde {f}=\left (\tilde {f}_{0},\ldots ,\tilde {f}_{n-1}\right )\) and
\({M_{l}^{T}}\) is the transpose of the matrix
Ml.
When
n,
d are integers with opposite parities as shown in (
10), only the first
κ and the last
κ elements of
\(\tilde {f}\) are nonzero. Also in the case when
n,
d are both odd integers, as can be seen in (
12), the first
m −
κ and the last
m −
κ − 2 elements of
\(\tilde {f}\) are zero. So we only use
n −
d + 1 columns of the Moore matrix in the encoding process.
In summary, the encoding of the optimal symmetric, alternating and Hermitian d-codes relies on converting the codewords of those codes to simplified linearized polynomials L(x) under carefully-chosen base of the extension fields, which enables us to treat encoding of those codes as evaluations of L(x) at linearly independent points.