The Korf Blog

This little chart made a few people apoplectic. They've never seen anything like it, and were (I guess still are) certain that this is either a fabrication or an error. Direct drives are phase locked, their speed control is essentially perfect, "hunting" is a myth, and so on and so forth.

Not only armchair thinkers fell into that trap. People who did their work and made their own measurements also insisted this is not happening. Why?

Their thinking was straightforward. The platter is rotating at about 0.5 Hz. Its inertia is quite large, and in combination with the slow speed this makes it unwise to go looking for artefacts above, say, 200 Hz. 400 Hz sensor sampling rate should be enough. 3kHz sinewave wow and flutter test signal gives us an effective sampling rate of 1.5 kHz, a 4x reserve. If something is not visible with a 400-500 Hz sensor sampling rate, or on the wow and flutter audio signal anayses, it's not there.

Let's have a look, shall we? Here's a screenshot of our rotation measurement system. This time, we are feeding it from a cheap and readily available Vernier GDX-FOR wireless sensor. It can be polled at 625 Hz over Bluetooth. Its error is around 0.01 °/s/√Hz which seems ridiculously low for what we are trying to measure.

We are measuring a typical midrange 1980s two-phase direct drive turntable. Nothing is playing, the platter is just rotating idly. If the "hunting" and "Golden Gate Bridge" hypothesis is true, it would certainly give us this behaviour.

And the charts do show that the performance of this turntable is, indeed, quite awful. The polar chart is pear-shaped, the raw timing data is far from a pretty sinewave, but... the histogram is almost perfectly Gaussian. I may argue that these are two intersecting bell curves, but the visuals do not persuade.

What would happen if we switch to a more advanced sensor?

Here's the same turntable, measured with an advanced sensor. This time, the sampling frequency is 4 kHz. And, looking just at the polar chart, there's hardly any change — the little GDX-FOR captured it well.

But the histogram couldn't be more different. There's a classic "Golden Gate Bridge" with two pronounced momentary speed centers. It's still completely awful, but in a very different way.

What happened? A glance at the raw data window tells us. GDX-FOR's noise filled those bins, giving us a pretty Gaussian bell curve where none really was. And its 625 Hz sampling rate leaves no headroom for any meaningful noise reduction. Here, the internal processing of the high speed sensor brings the noise down to 0.003 °/s/√Hz.

By the way, analyzing the audio track from a 3 kHz wow & flutter test signal playback provides similar results. The FM-demodulated raw data is full of tonearm and cartridge artefacts. While not as neatly spread as GDX-FOR's Gaussian noise, they also contaminate the histogram beyond the point of usefulness. This is easily proven by changing tonearms and cartridges between measurements — the distribution changes accordingly.

Please subscribe to receive blog updates in your inbox!