Like what you see?
Subscribe to the Korf Blog!
- Be notified when new articles are posted
- Access members-only secret preview sales
- Get deals and discounts on Korf Audio products
Expect 1-4 emails a month. We will never ever spam you. You can always unsubscribe—no hard feelings!
The Korf Blog
The inside story: our research,
development and opinions
development and opinions
3 December 2020
Any Point in Wow & Flutter? Part I
Wow, does time fly! It's been a month since our previous post. Thank you for your letters and comments about the ceramic spacer. We'll see if it can be brought to the market at an affordable price.
We have offered the limited number of headshells from the new batch to the blog subscribers, and they were gone in a day! I was especially happy to see the repeat customers. Thank you very much for your trust in Korf Audio, and I hope you would love your new headshells. They are now available to everyone at €199.00 plus shipping at our friends The Supersense.
We got a very interesting request in the meantime. A visitor to our lab asked: "Why do you keep a huge ancient Hewlett Packard 3561A signal analyzer? Surely you can do everything it does with a soundcard and some software?"
In my personal opinion, there are some things that the HP does a lot better. And below is an example of how we use both the hardware signal analyzer and today's software to investigate.
We got a very interesting request in the meantime. A visitor to our lab asked: "Why do you keep a huge ancient Hewlett Packard 3561A signal analyzer? Surely you can do everything it does with a soundcard and some software?"
In my personal opinion, there are some things that the HP does a lot better. And below is an example of how we use both the hardware signal analyzer and today's software to investigate.
The Wow and Flutter Story
Practically as soon as the standard playback speed of the recorded disc went from 78 rpm to 33 or 45 rpm, engineers understood that even the tiniest variation in the momentary rotation speed of the turntable is very audible. Unfortunately, back then the stroboscope disc was pretty much the only option available to check the said speed.
With the spread of FM (frequency modulation) radio, a bright idea came about: what if we play back a recording of a fixed frequency sine wave, and demodulate it with an FM decoder? Only the speed fluctuations would remain, and we could easily measure their magnitude! Thus, a wow and flutter meter was born.
Alas, in real life it wasn't quite so simple. The demodulated signal was dominated by the test disc's excentricity and the high frequency noise from the carrier tone harmonics. The resulting data was not representative of the turntable performance.
Undeterred, someone had an even brighter idea of applying the bandpass filter to the demodulated signal. This cut off most of the low and high frequency information, leaving just a narrow band between 1 and 100 Hz. This became known as a "weighted" wow and flutter measurement, and the sales people jumped at the "scientific" opportunity to show how their turntables were better than competition's.
Perceptual weighting of wow and flutter is not supported by reliable studies
To give some sort of respectability to this crude hack, an idea was proposed that "the human hearing is most sensitive to speed variation in this frequency range". Interestingly, the psychoacoustic study on the subject that everyone cites was done after the introduction of weighting in measuring equipment (Hisao Sakai, "Perceptibility of Wow and Flutter", JAES Volume 18 Issue 3 pp. 290-298; June 1970.) It cannot stand up to even cursory scrutiny, as Ampex's engineer John G. McKnight pointed out as soon as it was published. To me, the whole perception weighting idea feels like a post factum fabrication.
But the lure of reducing a complex behaviour to one easily measurable number was too strong. And, besides, there were no tools to extract more meaningful momentary speed data.
But the lure of reducing a complex behaviour to one easily measurable number was too strong. And, besides, there were no tools to extract more meaningful momentary speed data.
Enter Dynamic Signal Analyzers
By mid-1970s it became feasible to analyze the frequency content of signals in real time. Like an oscilloscope paints a clear picture of a signal in time domain, an analyzer gives you one in frequency domain: which frequencies exist in the signal, and what their magnitudes are.
For our momentary rotation speed investigation, the ability to plot a recorded tone in frequency domain allows us to sidestep the whole FM demodulation thing. We'll just look at the spectrum of a pure recorded sine signal. If it's nice and narrow, then our turntable drive is good. If it's broad, or asymmetric, or otherwise weird, then we have work to do.
For our momentary rotation speed investigation, the ability to plot a recorded tone in frequency domain allows us to sidestep the whole FM demodulation thing. We'll just look at the spectrum of a pure recorded sine signal. If it's nice and narrow, then our turntable drive is good. If it's broad, or asymmetric, or otherwise weird, then we have work to do.
Here's the spectrum of a 3150 Hz sine signal from a Dr Feickert test LP, played back by a quality turntable. I've captured it with a Virtins Multi Instrument software, but any other package that does FFT will do. That looks good, right?
Not so Fast
It only looks good because the resolution of our software signal analyzer is pretty bad. It's plotting 16384 points that are evenly spread over the signal's whole spectrum from 0 to 22500 Hz. This gives us a resolution of about 1.4 Hz, meaning smaller speed fluctuations would be missed. Which is quite unfortunate, because the platter rotates at about 0.55 Hz. A difference of 0.1 Hz is equivalent to 6 rpm.
We can increase the number of points, but that would unacceptably slow down our processing. Using 65536 points would mean analyzing the signal over a 3 second window. That's 1 1/2 turns of the platter. Might just as well read an average with a stroboscope instead.
Or we could concentrate those points over a narrower frequency range. Say, 200 Hz around the 3150 Hz signal. Then even 512 points would give us 0.5 Hz resolution. Unfortunately, no FFT software I know can do this without programming. But an HP 3561A can.
We can increase the number of points, but that would unacceptably slow down our processing. Using 65536 points would mean analyzing the signal over a 3 second window. That's 1 1/2 turns of the platter. Might just as well read an average with a stroboscope instead.
Or we could concentrate those points over a narrower frequency range. Say, 200 Hz around the 3150 Hz signal. Then even 512 points would give us 0.5 Hz resolution. Unfortunately, no FFT software I know can do this without programming. But an HP 3561A can.
The spectrum doesn't look so neat now, does it? Averaged over 30 measurements, it's more like a plateau than a peak. This is more or less what I expected, given HP's resolution and the test LP excentricity.
Can the performance of a turntable be estimated by test tone playback?
Have we found the Holy Grail? Is this the way to objectively assess real turntable performance? In the next part, we'll see the performance of different turntables on the HP 3561A screen.
We'll also do the measurements with a different LP, and answer the main question: can the momentary (as opposed to average) speed performance of a turntable be reliably estimated by playing back the test tone?
We'll also do the measurements with a different LP, and answer the main question: can the momentary (as opposed to average) speed performance of a turntable be reliably estimated by playing back the test tone?
Please subscribe to receive blog updates in your inbox!