OddEEC: Error Estimating Coding for Wireless Networks

• OddEEC is an error estimating coding scheme that repurposes Odd Sketch using bit sampling and maximum likelihood estimation to accurately measure Hamming distance in wireless packets.
• It mitigates sketch saturation by sampling a subset of bits, allowing the estimator to remain stable even in high bit-error-rate regimes.
• Empirical results show that OddEEC achieves comparable accuracy to gEEC and mEEC while dramatically reducing decoding complexity and lookup table size.

Searching arXiv for OddEEC and closely related work to ground the article in the relevant literature. OddEEC is an Error Estimating Coding (EEC) scheme for estimating the number of bit errors during packet transmission over wireless networks. It is introduced as “a novel EEC scheme” that adapts a data sketching technique named Odd Sketch to the EEC setting, with the adaptation centered on a bit sampling technique and a maximum likelihood estimator. In the reported experiments, OddEEC “overall achieves comparable estimation accuracy as competing schemes such as gEEC and mEEC, with much smaller decoding complexity” (Wang et al., 15 Aug 2025).

1. Definition and problem setting

Error estimating coding is described as “a standard technique for estimating the number of bit errors during packet transmission over wireless networks” (Wang et al., 15 Aug 2025). Within this setting, OddEEC treats packets as bit-strings $P$ and $P'$ of length $l$, and frames the target quantity as the Hamming distance between the transmitted and received packets. The paper states that this Hamming distance is equivalent to the symmetric difference cardinality between the sets corresponding to the indices of 1-bits in the two packets (Wang et al., 15 Aug 2025).

This construction places OddEEC at the intersection of wireless error estimation and compact sketching. The underlying sketching primitive is Odd Sketch, originally designed to estimate the Jaccard similarity between two sets $A$ and $B$. In Odd Sketch, a hash function $h: U \to [n]$ assigns elements to positions in an $n$-bit sketch, and bit $i$ is set to 1 iff the set of elements assigned to $i$ has odd size. A key inherited property is that the bitwise XOR $S_A \oplus S_B$ equals the Odd sketch of the symmetric difference set $P'$0 (Wang et al., 15 Aug 2025).

In OddEEC, this property is repurposed for error estimation. The transmitter forms a compact codeword from the packet, appends it to the packet, and the receiver compares the codeword derived from the received packet against the appended one. This suggests a design in which the algebra of symmetric difference is used as a proxy for bit-error counting rather than for set-similarity estimation.

2. Core construction and estimator

The OddEEC workflow is divided into encoding and decoding stages. On the sender side, the packet $P'$1 is converted to a set $P'$2 consisting of the indices of 1-bits, and the Odd sketch $P'$3 is computed and appended as a compact codeword. On the receiver side, the received packet $P'$4 is converted to a set $P'$5, the receiver computes $P'$6, forms the XOR $P'$7, and counts the number $P'$8 of 1-bits in that result (Wang et al., 15 Aug 2025).

The paper models $P'$9 as a binomial random variable: $l$0 with

$l$1

where $l$2 is the sketch length and $l$3 is the symmetric difference cardinality, equivalently the number of bit errors (Wang et al., 15 Aug 2025). From this model, OddEEC uses a method-of-moments estimator, described “with immunity,”

$l$4

This estimator is central to the scheme’s baseline decoding logic. It converts the observed parity-sketch disagreement count into an estimate of the underlying error count. The formulation also makes the saturation behavior explicit: as $l$5, the logarithm becomes unstable or undefined, limiting the usable error range for a fixed sketch length (Wang et al., 15 Aug 2025).

A plausible implication is that OddEEC’s practical effectiveness depends not only on sketch compactness but also on how well the system keeps the operating point away from this saturation regime.

3. Bit sampling and saturation control

The paper identifies saturation as a major obstacle for applying a naive Odd sketch directly to long packets or high bit-error-rate (BER) regimes. For large $l$6, a fixed-length sketch cannot accurately estimate the error count, because the observable $l$7 approaches $l$8 and the estimator becomes unstable (Wang et al., 15 Aug 2025).

OddEEC addresses this using a bit sampling technique. Rather than sketching the full packet, it samples a subset of bits of size

$l$9

where $A$0 is the sampling rate. Odd sketches are then constructed from the sampled packets. Under this procedure, the Hamming distance after sampling is reduced to

$A$1

and the estimator is rescaled as

$A$2

(Wang et al., 15 Aug 2025).

The paper further reports an approximate variance formula: $A$3 This is used for parameter tuning. In practice, the sampling rate is chosen so that

$A$4

“to keep the sketch unsaturated,” and it is tuned based on target packet lengths and anticipated BER (Wang et al., 15 Aug 2025).

This part of the construction is the principal mechanism by which OddEEC extends Odd Sketch to the EEC regime. The paper explicitly states that “the new bit sampling scheme replaces Broder’s embedding (which only preserves Jaccard similarity) to directly support Hamming distance estimation” (Wang et al., 15 Aug 2025). That statement is both methodological and conceptual: OddEEC is not merely reusing a sketch, but reengineering its input representation so that the relevant statistic is preserved for wireless error estimation.

4. Maximum likelihood estimation under codeword corruption

A defining complication in EEC is that the appended codeword may itself be corrupted during transmission. The paper highlights this as a challenge absent from the original Odd Sketch setting and introduces a maximum likelihood estimator (MLE) to address it (Wang et al., 15 Aug 2025).

Let $A$5 denote the number of 1s in

$A$6

where $A$7 is the corrupted transmitted codeword. For an observed value $A$8, the MLE estimates the BER $A$9 by

$B$0

The likelihood is written as a convolution: $B$1 where

$B$2

and

$B$3

The paper states that this estimator “can be efficiently implemented as a small lookup table (for $B$4)” (Wang et al., 15 Aug 2025).

The significance of this design is practical as well as statistical. The paper states that OddEEC is “unique among modern EEC schemes in its robustness to codeword corruption,” while also noting that “only OddEEC and gEEC can handle codeword corruption at all; mEEC fails under this scenario” (Wang et al., 15 Aug 2025). It further specifies that this robustness requires “only that the error rates for data and OddEEC bits are matched,” which “can be arranged by mixing codeword bits into the payload” (Wang et al., 15 Aug 2025).

This suggests that OddEEC’s estimator is tailored not simply to packet errors in the data field, but to an end-to-end transmission model in which the auxiliary coding bits are exposed to the same channel.

5. Multi-resolution design

The paper argues that “no single sampling scale (bit sampling parameter) suffices to cover the entire range of BERs of interest” (Wang et al., 15 Aug 2025). The stated trade-off is that for small BER, a larger sample size reduces quantization error, whereas for large BER, a smaller sample size helps avoid sketch saturation.

To address this, the paper proposes Multi-Resolution OddEEC. The codeword is divided into “two (or more) sub-sketches,” each using a different sampling length or resolution $B$5. Each sub-sketch is decoded independently using the MLE corresponding to its own corrupted count $B$6, and the joint likelihood is obtained as the product of likelihoods, yielding a global MLE for the tuple $B$7 (Wang et al., 15 Aug 2025).

The implementation strategy is explicitly table-driven. The paper states that “all possible combinations of $B$8 are precomputed and stored in a small table,” and gives the example that “for $B$9 per subcode, $h: U \to [n]$0 entries” suffice, enabling “instant lookup of the best $h: U \to [n]$1.” It summarizes the resulting decoding performance as “trivially fast (tens of nanoseconds)” (Wang et al., 15 Aug 2025).

This multi-resolution construction is presented as a way to “stitch together” multiple independently tuned subcodes for wide BER-range coverage (Wang et al., 15 Aug 2025). A plausible implication is that the method can be viewed as a composite estimator whose components specialize to different error regimes while sharing the same parity-sketching framework.

6. Empirical comparison with gEEC and mEEC

The paper compares OddEEC against gEEC and mEEC along two principal axes: estimation accuracy and decoding complexity (Wang et al., 15 Aug 2025). The most important reported result is that OddEEC “overall achieves comparable estimation accuracy as competing schemes such as gEEC and mEEC, with much smaller decoding complexity” (Wang et al., 15 Aug 2025).

For estimation accuracy, the paper distinguishes between single-resolution and two-resolution variants. It reports that Single-Resolution OddEEC “outperforms gEEC in certain BER ranges,” giving examples such as $h: U \to [n]$2 for smaller $h: U \to [n]$3 and $h: U \to [n]$4 for larger $h: U \to [n]$5, with relative MSE reduced “by factors of up to 2.5.” It also notes that a single-resolution design is worse than gEEC outside the range for which its parameter is tuned (Wang et al., 15 Aug 2025). By contrast, Two-Resolution OddEEC combines small and large $h: U \to [n]$6 and is reported to have relative MSE and log-MSE “as good or better than gEEC throughout the full range of interest,” while in the evaluated scenarios it is “always within 1.02-1.93$h: U \to [n]$7 the gEEC $h: U \to [n]$8, and often better in practice (factors between 1.2 and 2)” (Wang et al., 15 Aug 2025).

Against mEEC, the paper states that OddEEC achieves “essentially the same estimation accuracy as mEEC on codeword-immune (ideal) channel models,” and that “in terms of log-MSE, OddEEC is as good or better across almost all BERs” (Wang et al., 15 Aug 2025).

On decoding complexity, the contrast is sharper. The paper states that gEEC’s MLE “depends on the full state (all 96 bits), making the precomputed table intractably large ($h: U \to [n]$9 entries),” so “decoding must be done by a CPU in real time—for every packet—which takes tens of milliseconds.” It adds that this “fails real-time requirements,” citing as an example Wi-Fi 6 transmission time of $n$0 microseconds for a $n$1-byte packet (Wang et al., 15 Aug 2025). For mEEC, the paper states that the required precomputed table size is “~20 MB for typical settings,” and that mEEC “is not robust to codeword (subcode) corruption” (Wang et al., 15 Aug 2025). OddEEC, by contrast, “needs only a small, manageable lookup table (a few kilobytes)” and “enables decoding, by lookup, in tens of nanoseconds” (Wang et al., 15 Aug 2025).

The following summary reflects the comparative table reported in the source:

Scheme	Robust to Codeword Corruption	Decoding Latency
OddEEC	Yes	nanoseconds
gEEC	Yes	ms
mEEC	No	ns (if immune)

The paper also characterizes the table-size contrast as follows: OddEEC requires kB-scale precomputation, gEEC has an infeasible precomputed table, and mEEC requires MB-scale tables (Wang et al., 15 Aug 2025).

7. Novelty, scope, and interpretation

The paper identifies several elements as constituting OddEEC’s novelty. First, it is “a nontrivial adaptation” of Odd Sketch to EEC, with the adaptation explicitly tied to “its bit sampling technique and maximum likelihood estimator” (Wang et al., 15 Aug 2025). Second, the work extends Odd Sketch from symmetric difference cardinality estimation under codeword immunity to “the practical EEC problem (potentially corrupted codeword) via MLE” (Wang et al., 15 Aug 2025). Third, the multi-resolution approach combines multiple subcodes so as to cover a wider BER range than a single parameterization can handle (Wang et al., 15 Aug 2025).

From the reported results, the most stable characterization is that OddEEC prioritizes a balance of three properties: comparable estimation accuracy, robustness to codeword corruption, and very low decoding complexity. The paper repeatedly emphasizes that the reduction in runtime and table size is not achieved by discarding probabilistic decoding, but by choosing a sketch representation and likelihood structure whose sufficient statistics are compact enough for precomputation (Wang et al., 15 Aug 2025).

A common misconception would be to view OddEEC as merely a direct application of Odd Sketch. The paper argues otherwise: the original Odd Sketch is designed for Jaccard similarity, whereas OddEEC requires direct support for Hamming distance estimation and resilience to corruption of the appended codeword itself (Wang et al., 15 Aug 2025). Another potential misconception would be to treat its performance as uniformly superior to all alternatives in every regime. The reported results are more specific: single-resolution OddEEC is parameter-sensitive and can be worse than gEEC outside tuned regions, while the two-resolution design is used to recover broad-range performance (Wang et al., 15 Aug 2025).

In that sense, OddEEC is best understood as a sketch-based EEC framework rather than as a single fixed estimator. Its key components are the Odd-sketch representation, sampling-based control of saturation, and MLE-based decoding under corrupted codewords. Within the reported evaluation, this combination yields “comparable estimation accuracy” to gEEC and mEEC together with “much smaller decoding complexity” (Wang et al., 15 Aug 2025).