Three layers build the sound: a low-pass noise floor, a slowly drifting tone, and a reverb tail that ties them together.

The noise floor

The lowest layer is noise shaped to stay in the sub-bass register — below about 250 Hz. The technique is stateless: instead of filtering, the randomness itself is generated slowly, with each new value smoothly blended into the last. The result is a soft rumble with no high-frequency edge, felt more than heard.

The drifting tone

A sine wave sits in the bass register and slides through a range of roughly an octave over about a minute, fast enough that the sound is never static, slow enough that it doesn’t feel like melody. A second harmonic runs above it with a slightly different phase in each ear, giving the tone a gentle stereo width that the noise floor doesn’t have on its own.

The reverb

The reverb is where the two layers meet. Four delay lines at different lengths pick up both the noise and the tone from the audio history, blend them, and push the result through two diffusion stages. As the tone drifts in pitch, the reflections shift with it — that shifting quality is where most of the sense of movement comes from. Without the reverb, the noise and tone would sit side by side without much connection; the reverb smears them into a shared space.

The ring buffer

All three layers depend on the same piece of infrastructure: a texture on the GPU that holds about 1.4 seconds of audio history. The pink coloring of the noise and the reverb both read from this history simultaneously, which is why a texture is needed at all. The dry synthesis pass writes new audio into the texture each tick; the processing pass reads back as far as it needs and produces the final output.