[AUD-009][High] Reject unreconstructable DWT dimensions and use synthesis filters

[gabrielsantosphilips]

Audit finding

• ID: AUD-009
• Status: Verified defect
• Severity: High
• Confidence: High
• Audited revision: 2f479320d805a1f9f35ebe4afaaeeded48913a94

Problem

The DWT requires positive N and Levels but does not require N to be divisible by 2^Levels. It also exposes synthesis filters but SynthesisStep() reads the analysis filters instead.

Source:

numerical-toolbox-cpp/numerical/analysis/DiscreteWaveletTransform.hpp

Lines 81 to 86 in 2f47932

	{
	static_assert(std::is_floating_point_v<T>, "DiscreteWaveletTransform supports floating-point types only");
	static_assert(N > 0, "DiscreteWaveletTransform N must be > 0");
	static_assert(Levels > 0, "DiscreteWaveletTransform Levels must be > 0");
	static_assert(Taps >= 2, "DiscreteWaveletTransform Taps must be >= 2");

and

numerical-toolbox-cpp/numerical/analysis/DiscreteWaveletTransform.hpp

Lines 159 to 181 in 2f47932

	{
	T sumA{ 0 };
	T sumD{ 0 };
	for (std::size_t i = 0; i < half; ++i)
	{
	std::size_t fIdx = (n + outLen - 2 * i) % outLen;
	if (fIdx < Taps)
	{
	sumA += filters.lowAnalysis[fIdx] * ca[i];
	sumD += filters.highAnalysis[fIdx] * cd[i];
	}
	}
	out[n] = sumA + sumD;
	}
	}
	template<typename T, std::size_t N, std::size_t Levels, std::size_t Taps>
	OPTIMIZE_FOR_SPEED void DiscreteWaveletTransform<T, N, Levels, Taps>::Forward(const Signal& x, Signal& coeffs)
	{
	coeffs.clear();
	coeffs.resize(N, T{ 0 });
	for (std::size_t i = 0; i < N; ++i)

Reproduction

DiscreteWaveletTransform<float, 3, 1, 2> constructs and performs forward transform, then inverse reconstruction aborts on an internal bounded-vector access.

Acceptance criteria

• Enforce compatible N/Levels at compile time.
• Check exact runtime signal and coefficient sizes.
• Use the supplied synthesis filters, or narrow the API to orthogonal analysis-only coefficients.
• Add invalid-dimension compile checks and independent biorthogonal reconstruction tests.