Thermal noise — the random voltage fluctuations across a resistor — encodes the Boltzmann constant. Shot noise — the discrete current fluctuations through a junction — encodes the elementary charge. Both are fundamental constants hidden in apparently random signals. Measuring them requires hearing a whisper in a scream: the signal of interest is far smaller than the background noise of the measurement apparatus itself.
Peng, Zheng, and Yue solve this with cross-correlation. Two independent measurement channels record the same noisy signal simultaneously. The real signal — thermal or shot noise from the component under test — is correlated between channels. The background noise of each amplifier is independent. Cross-correlating the two channels preserves the correlated part and destroys the uncorrelated part. The whisper emerges from the scream not by making the microphone quieter, but by using two microphones and keeping only what they agree on.
The apparatus costs almost nothing: an ADC module, a low-noise op-amp, standard resistors. It achieves 1% accuracy on both k_B and e — good enough to see the physics, cheap enough for a high school lab. The measurement works because the statistical technique is more powerful than the hardware. You don't need a precision instrument when you have a precision method.
The structural point: the information was always in the noise. Both channels contain it. But in each channel individually, the signal is buried. Only the agreement between channels reveals the physical constant. This is the general principle: sometimes the information exists in every observation but is invisible in each one alone. It lives in the correlation, not in the data. Cross-correlation doesn't amplify the signal — it annihilates everything else.