Due to the fixed distribution of quantization bits, the codec cannot “transfer the bits” to frequencies that need them most. Unlike SBC, aptX will not “cut off” frequencies, but will add quantization noise to them, reducing the dynamic range of the audio.
We should not assume that using, for example, 2 bits for a band reduces the dynamic range to 12 dB: ADPCM allows up to 96 dB of dynamic range to be used, even with 2 quantization bits, but only with a certain type of signal.
ADPCM stores the difference between the current and the next value in numerical representation, instead of using the absolute value, as in PCM. This reduces the requirements for the number of bits needed to store the same (without loss) or almost the same (with relatively small rounding error) information. To reduce rounding errors, factor tables are applied.
When creating the codec, the authors calculated ADPCM coefficients on a set of music audio files. The closer audio signal is to the set of music on which the tables were built, the less quantization errors (noise) are generated by aptX.
Because of this, synthetic tests will always produce worse results than music.