Systematic MDS Codes: Theory & Constructions

• Systematic MDS codes are linear error-correcting codes with explicit generator matrices [I_k | A] that achieve the maximum minimum distance for given parameters.
• They ensure optimal erasure correction by enforcing that every square submatrix of the parity block is nonsingular, which is crucial for reliability.
• Recent developments introduce non-GRS constructions, including twisted and 'broken Cauchy' methods, expanding practical encoding and convolutional code applications.

Systematic maximum distance separable (MDS) codes are linear error-correcting codes that achieve the largest possible minimum distance for a given length and dimension, with the additional property that the original information symbols appear explicitly among the codeword symbols. This ensures optimal erasure correction capability and practical encoding/decoding efficiency. The systematic form is realized when the generator matrix is partitioned as $[I_k\mid A]$, where $I_k$ is the $k\times k$ identity and $A$ is a $k\times(n-k)$ parity block. Historically, generalized Reed–Solomon (GRS) codes have served as the canonical class of MDS codes. However, recent work has revealed infinite families of non-GRS MDS codes with explicit systematic generator matrices and efficient constructions, thus expanding the landscape of systematic MDS codes far beyond the GRS paradigm (Liu et al., 28 Jul 2025), and extending to convolutional regime (Barbero et al., 2017).

1. Foundations of Systematic MDS Codes

Systematic MDS codes over a finite field $\mathbb{F}_q$ (or its extension $\mathbb{F}_{q^d}$) are parameterized by length $n$, dimension $k$, and minimum distance $d=n-k+1$. In systematic form, a generator matrix $I_k$0 has the structure $I_k$1 for some $I_k$2 matrix $I_k$3. The code is MDS if and only if every $I_k$4 submatrix of $I_k$5 (with $I_k$6) is nonsingular [(Liu et al., 28 Jul 2025), Roth]. The most studied systematic MDS codes are GRS codes, for which $I_k$7 is a Cauchy matrix. By explicit construction, $I_k$8 generates a GRS code if $I_k$9 has all minors nonzero and admits a representation in terms of field locators and multipliers.

For convolutional codes, the systematic generator matrix is $k\times k$0, with $k\times k$1 a vector of polynomials. The code is called MDS if, for memory $k\times k$2, its column distance profile (CDP) achieves $k\times k$3, maximizing erasure recovery and free distance (Barbero et al., 2017).

2. Non-GRS Systematic MDS Codes: Twisted and Generalized Twisted RS Codes

Classical results showed the existence of non-GRS MDS codes, notably the Roth–Lempel construction. In recent developments, Beelen et al. introduced twisted Reed–Solomon (TRS) codes, and subsequent generalization led to generalized twisted Reed–Solomon (GTRS) codes. Two infinite families of non-GRS, systematic MDS codes were constructed using GTRS codes (Liu et al., 28 Jul 2025):

• Family 1: $k\times k$4 Require $k\times k$5. Code is over $k\times k$6 with locators $k\times k$7, nonzero multipliers $k\times k$8, and twist parameter $k\times k$9 such that $A$0 (with $A$1 a function of locators). Proven non-GRS for the full parameter range by Schur-square arguments.
• Family 2: $A$2 Require all $A$3 and $A$4 (or up to $A$5 if all locators are nonzero). Also non-GRS for stated parameter ranges.

In both cases, the non-GRS property is established by demonstrating that the Schur square (or its dual) of the code exceeds dimension $A$6, precluding GRS structure.

3. Systematic Generator Matrices for GTRS Codes

For one-twist GTRS codes, the standard generator matrix $A$7 evaluating $A$8 at the locators is systematically transformed to $A$9. Let $k\times(n-k)$0 be the $k\times(n-k)$1 Vandermonde matrix on the first $k\times(n-k)$2 locators, $k\times(n-k)$3 the GRS Cauchy-type matrix, and $k\times(n-k)$4, where $k\times(n-k)$5 is the $k\times(n-k)$6st unit vector. Parity block $k\times(n-k)$7 is given by

$k\times(n-k)$8

where $k\times(n-k)$9 and $\mathbb{F}_q$0 is the relevant interpolation polynomial. The block $\mathbb{F}_q$1 is not Cauchy in general, so $\mathbb{F}_q$2 cannot define a GRS code [(Liu et al., 28 Jul 2025), Prop. 4.2].

Code Family	Parity Block Structure	GRS/Non-GRS
GRS	Pure Cauchy matrix	Always GRS
GTRS	$\mathbb{F}_q$3	Non-GRS (if $\mathbb{F}_q$4)

This construction ensures systematic encoding and guarantees the MDS property provided $\mathbb{F}_q$5.

4. Systematic MDS Codes by “Breaking” the Cauchy Matrix

A new family of systematic non-GRS MDS codes may be constructed by perturbing a Cauchy matrix. Given $\mathbb{F}_q$6 and a $\mathbb{F}_q$7 Cauchy matrix $\mathbb{F}_q$8 over $\mathbb{F}_q$9, extend to $\mathbb{F}_{q^d}$0 and select $\mathbb{F}_{q^d}$1. Let $\mathbb{F}_{q^d}$2 denote the $\mathbb{F}_{q^d}$3 matrix with a single $\mathbb{F}_{q^d}$4 in the $\mathbb{F}_{q^d}$5 position and zeros elsewhere. Define $\mathbb{F}_{q^d}$6, and let the generator matrix be $\mathbb{F}_{q^d}$7. This code is always MDS and not GRS, as the parity block $\mathbb{F}_{q^d}$8 cannot be represented in a Cauchy form over $\mathbb{F}_{q^d}$9 due to the presence of $n$0 [(Liu et al., 28 Jul 2025), Thm. 5.1]. The MDS property is certified by minor expansion: any minor involving the $n$1 entry is a sum of two nonzero terms in disjoint fields, so never vanishing.

A specific example over $n$2 with $n$3, $n$4 is explicitly computed, illustrating the construction and non-GRS property.

5. Systematic MDS Convolutional Codes

For convolutional codes (rate $n$5), a systematic code is MDS if its column distance profile is $n$6 for $n$7, where $n$8 is encoder memory. Algebraic constructions achieving free distances $n$9 ($k$0) and $k$1 ($k$2) are given for suitable rates and field sizes, and systematic generator matrices in polynomial and truncated-block forms are provided (Barbero et al., 2017). For higher free distance, a branch-and-bound computer search enforces $k$3-superregularity (all proper minors nonzero) of the relevant partial parity-check matrix. The existence and parameters of systematic MDS convolutional codes are thus controlled by minor conditions analogous to the block code case.

Maximum attainable free distances for various field sizes are tabulated, e.g., over $k$4, rates $k$5, $k$6, and $k$7 allow free distances $k$8, $k$9, $d=n-k+1$0 respectively (Barbero et al., 2017).

6. Key Theorems and Structural Criteria

• Roth’s Theorem: An $d=n-k+1$1 block code with systematic generator matrix $d=n-k+1$2 is MDS if and only if every $d=n-k+1$3 submatrix of $d=n-k+1$4 ($d=n-k+1$5) is nonsingular [(Liu et al., 28 Jul 2025), Roth].
• GTRS Non-GRS Criteria: $d=n-k+1$6 is non-GRS for $d=n-k+1$7; $d=n-k+1$8 is non-GRS for $d=n-k+1$9 (or $I_k$00 if all $I_k$01).
• Systematic GTRS Form: For a single twist, systematic form is $I_k$02 as above.
• “Broken Cauchy” Construction: If $I_k$03 is any Cauchy block, extend to $I_k$04, $I_k$05, then $I_k$06 generates a non-GRS MDS code over $I_k$07.

7. Practical Implementation and Verification

All the above constructions can be implemented in software/hardware by explicitly building generator matrices, verifying full-rank conditions (MDS via minors), and checking the Cauchy/non-Cauchy nature of the parity blocks. The systematic form is particularly advantageous for encoder/decoder design and for practical applications requiring explicit access to uncoded symbols. The “broken Cauchy” and GTRS constructions provide infinite families of MDS codes not equivalent to GRS under field or parameter transformations, diversifying available code families and supporting robust communication and storage solutions (Liu et al., 28 Jul 2025, Barbero et al., 2017).