⚛️ quantum physics

Bias-Aware BP Decoding of Quantum Codes via Directional Degeneracy

This paper introduces a bias-aware belief propagation decoding framework for quantum CSS codes that leverages directional degeneracy through anisotropic Tanner-graph weights and bias parameters to significantly reduce logical error rates under biased noise without altering the underlying code construction.

Original authors: Mohammad Rowshan

Published 2026-01-15

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Mohammad Rowshan

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to solve a giant, complex puzzle to fix a broken machine. In the world of quantum computers, this "machine" is a fragile quantum state, and the "puzzle" is figuring out what went wrong so you can fix it without breaking anything else.

This paper introduces a smarter way to solve that puzzle by paying attention to the direction of the errors, rather than just counting how many there are.

Here is the breakdown of the paper's ideas using simple analogies:

1. The Problem: The "Noisy Room" and the "Ghost Errors"

Quantum computers are like rooms where the wind blows in specific patterns. Sometimes the wind blows hard from the North, but gently from the South. In technical terms, this is called anisotropic noise (noise that has a preferred direction).

When an error happens, it leaves a "syndrome" (a clue). But here is the tricky part: in quantum codes, many different mistakes can leave the exact same clue. This is called degeneracy.

The Analogy: Imagine you hear a crash in the kitchen. It could be a dropped plate, a falling book, or a cat knocking over a vase. All three leave the same "clue" (the noise). A standard decoder looks at the clue and says, "I don't know which one it was, so I'll guess randomly among all possibilities." This is inefficient and often leads to mistakes.

2. The Solution: Giving the Decoder a "Compass"

The authors propose a new method called Bias-Aware BP Decoding. Instead of treating every possible error as equally likely, they give the decoder a "compass" based on the hardware's layout.

The Map: They draw a map (called a Tanner graph) of the quantum computer.
The Weights: They put "weights" on the connections in this map. If the hardware is known to be more prone to errors in a specific direction (like a long row of wires), they mark those paths as "heavier" or more suspicious.
The Compass (Parameter $\beta$ ): They use a single control knob, called $\beta$ $β$ .
- If you turn the knob to zero, the decoder ignores the direction and guesses randomly (the old way).
- If you turn the knob up, the decoder starts saying, "Ah, errors are much more likely to happen on the 'heavy' paths. I will bet on those."

3. How It Works: The "Directional Cost"

In the old way, the decoder counts errors like steps: "One error here, one error there = 2 steps."
In this new way, the decoder counts directional cost:

If an error happens on a "heavy" path (where the hardware is weak), it counts as a big cost.
If an error happens on a "light" path, it counts as a small cost.

The decoder then looks at all the possible "ghost errors" (the degenerate classes) and asks: "Which group of errors has the lowest total directional cost?" It picks that group to fix.

4. The Results: A Much Cleaner Fix

The authors tested this on two types of quantum codes (the "Toric code" and the "NE3N code").

The Finding: By simply turning up the "directional knob" ( $\beta$ ) to match the hardware's natural weaknesses, they reduced the number of logical errors (the final mistakes that ruin the calculation) by 10 to 100 times.
The Catch: They didn't have to rebuild the computer or change the puzzle rules. They just changed how the decoder thinks about the clues.

5. The Warning: Don't Guess Wrong

The paper notes a crucial limitation: This only works if your "compass" is accurate.

The Analogy: If you tell the decoder, "Errors always come from the North," but the wind actually blows from the East, the decoder will make worse guesses than if it had no compass at all.
The method works best when the hardware actually has a clear direction (like a long strip of chips) and the decoder is tuned to match it.

Summary

Think of this paper as teaching a detective to look at the wind direction before solving a crime.

Old Way: "Someone broke a window. It could be a rock, a ball, or a bird. I'll guess."
New Way: "It's a windy day blowing from the East. The window is on the East side. It's highly likely a rock was thrown from that direction. I'll guess rock."

By using this "directional" logic, the quantum computer can fix its own mistakes much more accurately, without needing any new hardware or complex code changes.

Technical Summary: Bias-Aware BP Decoding of Quantum Codes via Directional Degeneracy

Problem Statement
Quantum Low-Density Parity-Check (qLDPC) codes, including lifted-product and quantum Tanner codes, offer asymptotically good parameters but face practical decoding challenges at finite blocklengths. These challenges stem from short cycles in Tanner graphs and degeneracy—the phenomenon where distinct physical errors share the same syndrome and act identically on the codespace. While physical noise is often anisotropic (e.g., dephasing dominating bit flips, or hardware layouts favoring specific coupling directions), existing decoding strategies typically either modify the code construction to match the channel or use decoders that fail to explicitly encode the directional structure of the Tanner graph. This work addresses the decoding side, aiming to leverage hardware-aware anisotropy without altering the underlying code.

Methodology
The paper proposes a framework that annotates the Tanner graphs of CSS codes with directional edge weights to reflect layout, scheduling, or calibrated noise anisotropy. The methodology proceeds through the following steps:

Directional Annotation: The Tanner graphs of the parity-check matrices ( $H_X, H_Z$ ) are augmented with non-negative edge weight matrices ( $D_X, D_Z$ ). Larger weights indicate stronger coupling along a preferred direction.
Per-Qubit Weights: Edge weights incident to each qubit are aggregated to form a per-qubit directional weight vector $w$ . This defines a directional cost $\Delta_w(E) = \langle w, E \rangle$ , which acts as a weighted Hamming weight where errors on "directionally important" qubits incur higher costs.
Directional Degeneracy Enumerator: The authors define a metric $\Gamma_X(s_Z; \beta)$ that enumerates degenerate error classes for a given syndrome, weighted by their directional cost. This enumerator summarizes how degeneracy concentrates along specific directions.
Anisotropic Decoding: A single bias parameter $\beta$ maps the directional weights into site-dependent log-likelihood ratios (LLRs). These LLRs serve as anisotropic priors for a standard Belief Propagation (BP) followed by Ordered Statistics Decoding (OSD) pipeline. The code construction and decoder architecture remain unchanged; only the priors and candidate scoring functions are modified.

Key Contributions
The paper makes three primary theoretical and practical contributions:

Formalization of Directional Degeneracy: The authors introduce a directional degeneracy enumerator and a class score (the minimum directional cost within a coset). They prove that directional weighting can "thin out" competitive low-cost error classes, reducing the decoder's confusion between logical errors.
Analytical Bounds:
- Distance Bounds: The paper derives bounds relating directional distances to standard Hamming distances, showing that mild anisotropy preserves distances up to a scaling factor.
- Degeneracy Reduction: A directionality-induced degeneracy bound is proven, explicitly quantifying how directional bias reduces the number of admissible degenerate error classes per syndrome as a function of code rate, distance, and bias strength.
- MacWilliams-Type Representation: A global enumerator is presented with a MacWilliams-type expression, connecting directional degeneracy to the dual code and allowing for factorization over coordinates in the dual domain.
Practical Decoder Implementation: The authors propose a Bias-Aware BP+OSD decoder. This decoder uses the directional weights to generate site-dependent priors via a parameter $\beta$ , allowing standard decoders to exploit hardware anisotropy without re-engineering the code.

Results
Finite-length simulations under code-capacity noise demonstrate the efficacy of the approach:

Logical Error Rate Reduction: For both the $[[162, 2, 9]]$ Toric code and the $[[36, 4]]$ NE3N planar code, the anisotropic decoder achieves significant reductions in logical error rates compared to isotropic baselines.
Magnitude of Gain: At moderate physical error rates, the method yields reductions of one to two orders of magnitude.
Tunability: The parameter $\beta$ acts as an effective control knob; simulations show that increasing $\beta$ (up to an optimal point) systematically improves performance, confirming that the directional prior effectively aligns with the geometric orientation of the lattice.

Significance and Claims
The paper claims that modest anisotropy, controlled by a single parameter $\beta$ , is a simple and effective route to hardware-aware decoding gains. The significance lies in the ability to improve decoding performance without changing the code construction or the fundamental decoder architecture. The framework is described as "portable," capable of being derived from geometry and scheduling data and applied to existing BP+OSD pipelines.

The authors maintain a modest stance, noting that anisotropic priors are not a universal solution ("silver bullet"). They acknowledge that for codes with essentially isotropic degeneracy (e.g., certain ensemble-style qLDPCs) or hardware with i.i.d. noise, the benefits may be negligible or negative if the assumed orientation is misaligned. Furthermore, they suggest that for complex cross-qubit correlations, richer models beyond simple directional fields may be required. The work establishes a theoretical foundation for directional degeneracy and provides a practical, low-overhead method to leverage known hardware biases in quantum error correction.