Abstract:We introduce two notions of discrepancy between binary vectors, which are not metric functions in general but nonetheless capture the mathematical structure of the binary asymmetric channel. In turn, these lead to two new fundamental parameters of binary error-correcting codes, both of which measure the probability that the maximum likelihood decoder fails. We then derive various bounds for the cardinality and weight distribution of a binary code in terms of these new parameters, giving examples of codes meeting the bounds with equality.
What problem does this paper attempt to address?
The main problem that this paper attempts to solve is related to the coding theory in the Binary Asymmetric Channel (BAC). Specifically, the author introduced two new discrepancy measures between binary vectors. Although these measures are not metric functions in the general sense, they can capture the mathematical structure of BAC. Based on these discrepancy measures, the author defined two new fundamental parameters of binary error - correcting codes, namely minimum discrepancy and minimum symmetric discrepancy respectively. These two parameters are used to measure the probability of maximum - likelihood decoder failure.
The main contributions of the paper include:
1. **Introduction of discrepancy measures**: Defined two discrepancy measures \(\delta(x, y)\) and \(\hat{\delta}(x, y)\), and studied their properties and their relationships with the traditional Hamming distance.
2. **Decoder performance analysis**: Proved that the maximum - likelihood decoder is equivalent to the minimum - discrepancy decoder in BAC, and gave the upper bounds of the maximum - likelihood decoder failure probability.
3. **Bounds of code size**: Derived the upper and lower bounds of the binary code size with a given minimum (symmetric) discrepancy by combinatorial methods.
4. **Connection with classical coding theory**: Showed how to apply the bounds in classical coding theory to codes under discrepancy measures, and provided specific code examples that satisfy these bounds.
### Specific problem analysis
1. **Problem background**:
- **Binary Asymmetric Channel**: In BAC, the probability \(p\) that 0 becomes 1 is usually less than the probability \(q\) that 1 becomes 0. This channel can be modeled as a discrete memoryless channel, where the noise acts independently on each component of the binary vector.
- **Neural coding**: BAC has important applications in the field of neural coding, especially in the study of discretized receptive field codes.
2. **Main contributions**:
- **Discrepancy measures**:
- Defined \(\delta(x, y)=\gamma d_{10}(y, x)+d_{01}(y, x)\), where \(\gamma = \log\left(\frac{q}{1 - p}\cdot\frac{1 - q}{p}\right)\).
- Defined \(\hat{\delta}(x, y)=\delta(x, y)-\omega_H(y)(\gamma - 1)\).
- **Decoder equivalence**: Proved that the maximum - likelihood decoder is equivalent to the minimum - discrepancy decoder.
- **Upper bounds of failure probability**: Gave two upper bounds of the maximum - likelihood decoder failure probability, based on \(\delta\) and \(\hat{\delta}\) respectively.
3. **Bounds of code size**:
- Derived the upper and lower bounds of the binary code size with a given minimum (symmetric) discrepancy by combinatorial methods.
- Compared these bounds with those in classical coding theory, and provided specific code examples that satisfy these bounds.
### Conclusion
This paper provides a new perspective for the coding theory in the binary asymmetric channel by introducing new discrepancy measures. These new measures not only help to understand the mathematical structure of BAC, but also provide a theoretical basis for designing more effective error - correcting codes. By analyzing the performance of the maximum - likelihood decoder and deriving the bounds of code size, the paper provides an important reference for future coding theory research.