A randomness extractor cleans imperfect noise before cryptographic use. Entropy sources may be biased, correlated, insufficient, predictable, or influenced by outside factors, so one source is not automatically trustworthy. section 8.1
A practical extraction pattern is:
- Gather noise from several sources, such as physical phenomena measured by hardware random number generators. section 8.1
- Hash the different inputs to transform their raw noise. section 8.1
- XOR the resulting digests together before using the output for cryptographic applications. section 8.1
Combining noisy measurements
Suppose one sensor produces a zero slightly more often than a one, while successive bits from another sensor are correlated. Hash both inputs and XOR the digests to gather and clean their noise. Extraction can be slow, however, making high-volume random-number generation a possible bottleneck. section 8.1

