Feng-Rao Majority Voting Decoding

• Feng-Rao majority voting is a decoding method that relies on well-behaving pairs and order domains to uniquely correct errors up to half the designed minimum distance.
• It leverages dual and primary code structures with combinatorial infrastructures such as basis products and syndrome computation for efficient error recovery.
• Improved bounds on minimum distance and generalized Hamming weights show enhanced code parameters, though they do not automatically translate to extended majority voting decoders.

Feng-Rao majority voting is a decoding method attached to the Feng-Rao bound and related order bounds in algebraic geometry, order domain, and evaluation-code settings. In its standard form, it uses information on well-behaving pairs to decode uniquely up to $$\left\lfloor \frac{d_{\mathrm{FR}}-1}{2}\right\rfloor$$ errors, where $d_{\mathrm{FR}}$ is the designed minimum distance given by the bound. Work on primary codes established that the Feng-Rao bound for dual codes and the analogous Andersen-Geil bound for primary codes are consequences of each other, implying that the Feng-Rao decoding algorithm can be applied to primary codes up to half their designed minimum distance (Geil et al., 2012). Subsequent papers sharpened the associated minimum-distance and generalized-Hamming-weight bounds, and also clarified that improved bounds do not automatically yield a corresponding majority voting decoder in every strengthened setting (Geil et al., 2013).

1. Historical placement and scope

The Feng-Rao bound was introduced as a strong lower bound for the minimum distance of dual algebraic-geometric codes and, together with that bound, came a decoding algorithm up to half the bound. Andersen and Geil later developed an analogous bound for primary codes, and the relation between the two viewpoints was made explicit by showing that the Feng-Rao bound for dual codes and Andersen and Geil’s bound for primary codes are consequences of each other (Geil et al., 2012).

This equivalence is operationally significant. It implies that the Feng-Rao decoding algorithm can be applied to decode primary codes up to half their designed minimum distance, and the technique applies to any linear code for which information on well-behaving pairs is available. The same source emphasizes that this gives efficient decoding for a large class of codes for which no non-trivial decoding algorithm was previously known, including important families of multivariate polynomial codes (Geil et al., 2012).

The literature represented here also places Feng-Rao majority voting inside a broader hierarchy of order-bound methods. For primary codes, Matsumoto and Miura derived from the Feng-Rao bound a bound for primary one-point algebraic geometric codes and showed how to decode up to what is guaranteed by their bound, while the primary-code formulation in (Geil et al., 2012) avoids the differential-based exposition associated with that earlier treatment. For dual codes, later work by Salazar, Dunn and Graham, and then further improvements in (Geil et al., 2013), extended the range of combinatorial structures used to sharpen the underlying distance estimates.

2. Combinatorial infrastructure: bases, products, and well-behaving pairs

The basic setup uses bases of $\mathbb{F}_q^n$. If $B=\{b_1,\ldots,b_n\}$ is a basis and $I\subseteq\{1,\ldots,n\}$, then the primary code is

$$C(B,I)=\operatorname{span}_{\mathbb{F}_q}\{b_i\mid i\in I\}.$$

With a second basis $U=\{u_1,\ldots,u_n\}$, the componentwise product is

$$(u_1,\ldots,u_n)*(v_1,\ldots,v_n)=(u_1v_1,\ldots,u_nv_n),$$

and if $L_i=\operatorname{span}\{b_1,\ldots,b_i\}$, then $\rho_B(v)=l$ when $d_{\mathrm{FR}}$0 (Geil et al., 2012).

An ordered pair $d_{\mathrm{FR}}$1 is well-behaving (WB) with respect to $d_{\mathrm{FR}}$2 if

$d_{\mathrm{FR}}$3

The same source also records the relaxed notions WWB and OWB. These combinatorial objects control both lower bounds on minimum distance and the majority voting mechanism itself (Geil et al., 2012).

For dual codes,

$d_{\mathrm{FR}}$4

where

$d_{\mathrm{FR}}$5

For primary codes,

$d_{\mathrm{FR}}$6

where

$d_{\mathrm{FR}}$7

The primary-code interpretation given in (Geil et al., 2012) is that the minimum distance is controlled by counting, for each index in the support, the number of WB pairs that “cover” it.

Further work on dual codes generalized this combinatorial infrastructure to three bases $d_{\mathrm{FR}}$8 and promoted one-way well-behaving pairs as the central relaxed notion. In that notation, $d_{\mathrm{FR}}$9 is OWB if

$\mathbb{F}_q^n$0

and (Geil et al., 2013) states that it demonstrates the advantage of working with one-way well-behaving pairs rather than weakly well-behaving or well-behaving pairs.

3. Majority voting decoding for primary and dual codes

The majority voting algorithm for primary codes in (Geil et al., 2012) is obtained by translating the primary-code problem into the dual-code syndrome framework. If a primary code $\mathbb{F}_q^n$1 is given and one receives

$\mathbb{F}_q^n$2

the first step is to compute a dual basis $\mathbb{F}_q^n$3 satisfying

$\mathbb{F}_q^n$4

Lemma 5, as summarized in the source, transfers WB-pair information from $\mathbb{F}_q^n$5 to $\mathbb{F}_q^n$6, so that the dual-code decoding machinery becomes available (Geil et al., 2012).

The syndrome stage is then split into known and unknown components. For positions corresponding to $\mathbb{F}_q^n$7, the syndrome components

$\mathbb{F}_q^n$8

are known, because $\mathbb{F}_q^n$9 for $B=\{b_1,\ldots,b_n\}$0. The remaining unknown syndrome values are recovered iteratively by majority voting. For each unknown syndrome $B=\{b_1,\ldots,b_n\}$1, one considers candidate WB pairs $B=\{b_1,\ldots,b_n\}$2 with $B=\{b_1,\ldots,b_n\}$3 and enough already known data; each such pair yields a candidate value for $B=\{b_1,\ldots,b_n\}$4 via the relevant linear constraint, and if the number of errors is up to half $B=\{b_1,\ldots,b_n\}$5, the majority vote yields the correct value (Geil et al., 2012).

After all syndromes are determined, the error vector is recovered by solving the resulting linear system. The source states that the algorithm can be implemented in $B=\{b_1,\ldots,b_n\}$6 operations, after a potentially more costly one-time preprocessing to compute bases and WB pairings. Structurally, the majority voting step is described there as nearly identical for primary and dual codes, owing to the duality via basis correspondence (Geil et al., 2012).

The same paper gives a concrete example with a primary code of dimension $B=\{b_1,\ldots,b_n\}$7 and index set $B=\{b_1,\ldots,b_n\}$8, where an error in one coordinate is recovered by the syndrome-majority mechanism. That example is presented as an instance of the general statement that unique decoding is possible up to

$B=\{b_1,\ldots,b_n\}$9

errors (Geil et al., 2012).

4. Improved bounds and their decoding implications

The 2013 paper on primary codes introduces a new bound for the minimum distance of a general primary linear code and states that for affine variety codes defined from generalised $I\subseteq\{1,\ldots,n\}$0 curves the new bound often improves dramatically on the Feng-Rao bound for primary codes. Its key innovations are case analysis on coefficient positions, the introduction of strongly one-way well-behaving (SOWB) pairs, and a footprint-method generalization based on unions of suitable supersets (Geil et al., 2013).

In the notation of that paper, if

$I\subseteq\{1,\ldots,n\}$1

and $I\subseteq\{1,\ldots,n\}$2 is chosen appropriately, then Theorem 3.1 is summarized as

$I\subseteq\{1,\ldots,n\}$3

For generalized $I\subseteq\{1,\ldots,n\}$4 curves, Theorem 4.3 is reported in the explicit form

$I\subseteq\{1,\ldots,n\}$5

where $I\subseteq\{1,\ldots,n\}$6 is a case-dependent additional term, described as exactly the improvement over the traditional Feng-Rao bound (Geil et al., 2013).

The decoding consequence is more limited than the distance improvement. The same source states that the Feng-Rao majority voting decoder corrects up to $I\subseteq\{1,\ldots,n\}$7 errors when $I\subseteq\{1,\ldots,n\}$8 comes from the original well-behaving version of the bound, but that majority voting decoding algorithms are only known for the WB version and do not exist for the SOWB or OWB settings. It therefore identifies as an open problem the construction of a decoding algorithm that can use the extra gain coming from SOWB or the more general footprint bound (Geil et al., 2013).

For dual codes, (Geil et al., 2013) presents a different strengthening trajectory. It states that the advisory bound of Salazar, Dunn and Graham and the new improvement are lifted so that they deal with generalized Hamming weights, and it also demonstrates the advantage of working with one-way well-behaving pairs rather than weakly well-behaving or well-behaving pairs. The minimum-distance version is summarized there as

$I\subseteq\{1,\ldots,n\}$9

A plausible implication is that strengthened order bounds and implementable majority voting decoders must be distinguished carefully: the primary-code SOWB setting in (Geil et al., 2013) explicitly lacks a corresponding majority voting algorithm, whereas (Geil et al., 2013) frames OWB-based improvements as directly relevant to the decoding radius of Feng-Rao-type majority voting for dual codes.

5. Generalized Hamming weights and Feng-Rao numbers

The majority voting paradigm is closely tied to generalized Feng-Rao distances and Feng-Rao numbers. For a numerical semigroup $$C(B,I)=\operatorname{span}_{\mathbb{F}_q}\{b_i\mid i\in I\}.$$0, (Delgado et al., 2011) defines

$$C(B,I)=\operatorname{span}_{\mathbb{F}_q}\{b_i\mid i\in I\}.$$1

with

$$C(B,I)=\operatorname{span}_{\mathbb{F}_q}\{b_i\mid i\in I\}.$$2

For large $$C(B,I)=\operatorname{span}_{\mathbb{F}_q}\{b_i\mid i\in I\}.$$3,

$$C(B,I)=\operatorname{span}_{\mathbb{F}_q}\{b_i\mid i\in I\}.$$4

where $$C(B,I)=\operatorname{span}_{\mathbb{F}_q}\{b_i\mid i\in I\}.$$5 is the $$C(B,I)=\operatorname{span}_{\mathbb{F}_q}\{b_i\mid i\in I\}.$$6-th Feng-Rao number. The same source states that these quantities are introduced as lower bounds on generalized Hamming weights for algebraic geometry codes and play a key role in Feng-Rao majority voting decoding (Delgado et al., 2011).

For semigroups generated by an interval,

$$C(B,I)=\operatorname{span}_{\mathbb{F}_q}\{b_i\mid i\in I\}.$$7

(Delgado et al., 2011) gives a closed formula for $$C(B,I)=\operatorname{span}_{\mathbb{F}_q}\{b_i\mid i\in I\}.$$8 and develops the structural theory of amenable sets and ordered amenable sets. Among the conclusions recorded in the source are that an optimal configuration is amenable if and only if it is closed under taking divisors above $$C(B,I)=\operatorname{span}_{\mathbb{F}_q}\{b_i\mid i\in I\}.$$9, and that ordered amenable sets minimize the number of divisors. The same paper states that the explicit computation of $U=\{u_1,\ldots,u_n\}$0 yields precise order bounds for code parameters and enables efficient decoding algorithms, because the majority voting threshold depends directly on these values (Delgado et al., 2011).

The extension to higher weights also appears in the later semigroup papers. For inductive numerical semigroups, the second Feng-Rao number is explicitly computed, and the source states that this number determines the asymptotical behaviour of the order bound for the second Hamming weight of one-point AG codes. In that setting,

$U=\{u_1,\ldots,u_n\}$1

and for $U=\{u_1,\ldots,u_n\}$2,

$U=\{u_1,\ldots,u_n\}$3

The same account adds that Feng-Rao decoding corrects up to half the order bound and that explicit knowledge of $U=\{u_1,\ldots,u_n\}$4 gives provable guarantees in this setting (Farrán et al., 2015).

For telescopic numerical semigroups, the central result is that the second Feng-Rao number agrees with the multiplicity: $U=\{u_1,\ldots,u_n\}$5 The same source connects this to the second generalized Hamming weight by

$U=\{u_1,\ldots,u_n\}$6

and states that this improves upon other estimates such as the Griesmer Order Bound (Farrán et al., 2016).

6. Scope, limitations, and recurrent misconceptions

A recurrent misconception is that Feng-Rao majority voting is confined to classical one-point AG codes. The primary-code formulation in (Geil et al., 2012) states instead that the technique applies to any linear code for which information on well-behaving pairs is available, and it explicitly lists important families of multivariate polynomial codes among the classes that thereby become efficiently decodable (Geil et al., 2012).

Another recurrent misconception is that every improvement of a Feng-Rao-type lower bound immediately enlarges the actual majority voting decoding radius. The primary-code improvement in (Geil et al., 2013) is explicit that, although the new bound can improve minimum distance and generalized Hamming weights dramatically when only order domain condition (C1) holds but not (C2), no majority voting decoding algorithms are known for the SOWB or OWB settings used there. The paper therefore separates improved code parameters from currently available decoding algorithms (Geil et al., 2013).

A third misconception is that higher Feng-Rao numbers are only asymptotic bookkeeping devices. The numerical-semigroup literature represented here treats them as operational invariants for generalized Hamming weights and order bounds. The interval-generated, inductive, and telescopic semigroup papers all connect explicit formulas for Feng-Rao numbers to lower bounds for higher Hamming weights and to decoding guarantees phrased in terms of order bounds and majority voting thresholds (Delgado et al., 2011).

Taken together, these results place Feng-Rao majority voting at the intersection of combinatorial pair structures, duality between primary and dual codes, and semigroup/order-bound asymptotics. Its classical form is algorithmically concrete and polynomial-time, its design radius is governed by WB-based lower bounds, and its modern developments show both the power and the current limits of extending majority logic beyond the original well-behaving framework (Geil et al., 2012).