New Belt, and a Wow & Flutter Deep-Dive

So I bought one of those new Rega ‘Reference’ belts for my old Planar 3. But even if you’re not one of the handful of Rega owners among SBAF, you might like to keep reading as this is mostly about measuring and visualizing wow & flutter, and evaluating different measurement techniques, than about the belt itself (especially the first two 2-3 posts). Conversely, if you DGAF about the details of the measurements and are mostly interested in the new belt, you can focus on the third 4th post.

TLDR: Using the Wow and Flutter Visualiser plug-in for Audacity in conjunction with the WFGUI software wow & flutter meter is more informative and probably more accurate than using phone-based apps, and almost as easy (but not as portable). And yes, the Rega Reference belt is pretty good, but it runs slower than some others - which might be fine if you're not using a PSU, as uncontrolled Regas usually run fast; or can be compensated for if you're using the speed-adjustable Neo PSU.

Background

Turntable belts can affect both absolute speed and variation about the average speed. The latter has historically been expressed in terms of wow (amount of <~5 Hz oscillation) and flutter (~30-100 Hz), but in modern parlance ‘wow & flutter’ is defined as variation centred around ~4 Hz (the middle curve of the 3 below), it having been found that this low frequency oscillation is subjectively the most annoying.



There are various standards for expressing the magnitude of wow & flutter, which can lead to confusion when assessing turntable specs (and amusement when viewing dick-waving exchanges on certain vinyl or tape forums). This is a good concise review that anybody really interested should read, but here it’s enough to say there are 3 different W&F parameters that are most often quoted (all with the 4 Hz weighting curve):

• Weighted- or Quasi-peak: weighted max amplitude deflection in the dataset, with higher weighting given to longer durations (wider peaks, up to a max at 10% of the cycle under analysis)
• 2sigma: the max amplitude deflection in a dataset that’s exceeded no more than 5% of the time
• RMS: the max root-mean-square amplitude deflection, over a 1s or 5s measurement cycle, in the dataset

All these have the 4 Hz weighting curve applied, but sometimes peak-to-peak (absolute max) and unweighted (no 4 Hz filter) numbers are also given. Typically, the 2σ numbers are a bit smaller than the quasi-peak numbers, and the RMS numbers are somewhat over half the quasi-peak. Here’s how the different quantities would look if the 4 Hz filtered W&F was a pure sine wave:



For IEC or DIN standard compliance the quasi-peak or 2σ numbers are usually quoted, whereas the JIS (Japanese) standard uses the RMS number. It’s this that can lead to confusion when comparing different turntable specs or measurements to each other, and when assessing what’s a “good” performance vs recommended standards – e.g. DIN-compliant ‘hi-fi’ gear is often talked about as having W&F better than 0.2% (quasi-peak) whereas JIS specs and even some amateur measurements for Japanese direct-drive turntables get down to a tenth of that (RMS).

Measuring

Phone-based apps use one of two ways to measure speed and W&F: some use the phone’s motion sensor to record the rotation while sitting directly on the platter (e.g., Turntable Speed), others (the now obsolete PlatterSpeed) use the phone’s microphone to record the variation of a fixed frequency signal played from a test record. There are pros and cons to both, the test-record method accounting for the drag of the stylus, the on-platter method avoiding possible artifacts introduced from the rest of the playback and recording chain (as discussed e.g. here). Historically, standard W&F measurement methods used test tapes or records with reference signals which were played back directly into specialized measurement instruments. There are now computer-based programs that can analyze recordings of a test track (recorded with a mic or captured directly into a computer’s sound card).

I used 4 different methods: PlatterSpeed, using my old (Android 5.0) Samsung S4 and its internal microphone; Turntable Speed on a Samsung S10; WFGUI, a free PC program; and the Wow and Flutter Visualiser plugin for Audacity. For the computer methods I used 44.1 kHz WAV files recorded with the (obsolete) Bosch iNVH app on my Galaxy S4 with a Dayton iMM-6 microphone. The input signal used for the computer methods and for PlatterSpeed was the 30 sec 3150 Hz track on the Tacet ‘Vinyl: Check’ test record. I measured 3 different belts: the Reference, my current Edwards Little Belter (aka Blue), and a standard Rega black belt which is older than the blue belt but hasn’t had as much use. But before comparing the belts to each other, it’s useful to evaluate the measurement characteristics of each app…

 

Comparing measurement apps

PlatterSpeed uses the phone’s mic to record the variations in a 3150 or 1000 Hz signal from a test record, and displays the mean frequency and raw and ‘notch filtered’ variance graphically on the screen:



Mean frequency can be converted to average speed by dividing it by 3150 (or 1000) and multiplying by 33.33 RPM (because your test record will be a 33 – do 45 RPM versions even exist?). I haven’t seen documented anywhere what ‘notch filtered’ actually means. There was a paid upgrade that quoted the ‘notch’ figures as IEC 2σ and also gave IEC ‘dynamically weighted’ numbers; I’d guess this latter to be 4 Hz weighted peak, and the free version’s ‘notch’ to be either quasi-peak or 2σ, but it’s frustrating not to know for sure.

The program ignores frequencies outside a certain window (maybe +/- 5%? Not sure) around the nominal test frequency so artifacts too far away from this won’t screw up the results, and the raw data can be saved to the phone as an .xml file that can be further (over) analyzed as I did here. Because of the low (10 Hz) sampling frequency though, the dataset is subject to aliasing and Fourier analysis is limited to 5 Hz and below; and because the ‘notch’ results are so dominated by the 0.56 Hz (331/3 RPM) oscillation in the raw data (I’m not certain the app is applying the 4 Hz weighting in a standards-compliant way) it’s hard to take them seriously. Feickert is not now selling its 7” test record nor updating the app. Even with a good test record, though, repeated tests show large variations in the results.

Turntable Speed (iOS version too) uses the phone’s motion sensors to record rotation speed and variation, and outputs average speed and it’s deviation from 331/3 or 45 RPM (yep, it works on 45); raw variation about the average speed; and 2σ and RMS wow. Although it’s designed to be used with the phone sitting directly on the platter, if your phone isn’t too huge you can include a record and put the arm down on one of the outer tracks so that the speed with the drag of the stylus (i.e. speed while actually playing) can be measured too:



The upside of measuring without the involvement of a test record is that possible artifacts in the playback and recording chain are avoided. However, the drag of the stylus isn’t included either, which could make a material difference to the speed (as in this example) so if you have a small enough phone it’s probably best to test with the arm down on a disc. Of course, the results from this method are dependent on the precision and accuracy of the phone’s motion sensors, which might be different in different phones.

The app’s results page can be saved for later comparison, but if the actual datasets are saved anywhere I haven’t yet managed to find them. The rudimentary graph plotted at the top suggests the sampling frequency is 10 Hz, which means, as with PlatterSpeed, there could be some bad aliasing; and combined with the short sample record lengths of 10 sec, I’m not sure how accurate the wow calculations can really be. Nonetheless, I like this app better than PlatterSpeed for ease and speed of use and better consistency of results over multiple trials (presuming that’s not reflecting imprecise motion sensing by my phone…).

WFGUI (Wow & Flutter Graphical User Interface) is a PC program that takes audio input from a computer soundcard and analyzes and displays it in a way that mimics old-school W&F meters:



Input can be real-time from a turntable playing a 3000 or 3150 Hz test track, or from a recording of such a track. I don’t have a laptop and being too lazy to lug my turntable, preamp and power supply downstairs to the computer, I used WAV files recorded on my phone. Sampling is done at 44.1 kHz and the program ignores any signal frequency outside +/- 5% of the chosen test frequency and any signal level below ~24 mV so spurious noise is avoided. Real-time frequency, quasi-peak, and RMS are displayed in terms of user-selectable DIN-weighted W&F, weighted wow, weighted flutter, or unweighted W&F as in the curves on the first graph in the previous post. There’s also a ‘peak hold’ display showing max RMS and max quasi-peak over the last 10s, and a fake oscilloscope showing a curve interpolated from the last 125 samples (280 ms). The % range of the meter and oscilloscope displays can be changed.

A log file of the 10s-averaged max quasi-peak, the instantaneous RMS results (not 10s max of the latter, despite what seems to be implied in the documentation) and the instantaneous frequency can be saved, but unfortunately this has only 1s resolution. If the signal is switched on while the program is sampling, the manual recommends throwing away the first 20s of the quasi-peak results because they might contain impulses. Looking at the logged results when starting the signal before the sampling, though, it seems that the first 20s of quasi-peak data are always the highest, so it’s probably best to discard them in this situation too. This is unfortunate, because many LP test tracks being only 30s long, not much record length remains. This is less of an issue for the RMS results because they’re calculated over a 1s period (DIN standard) so only the first second should be discarded.

Another caveat: because the saved files have much lower resolution than the sample rate, switching the meter on while the signal is playing means the actual samples used to calculate the numbers in the log file will be different every time. If you only have a short test record, it’d be best to log a few trials and take the maximum numbers (after discarding the first 20s of quasi-peak and 1s of RMS) as your final results. Similarly, to calculate average RPM from a short test record (30s at 33.33 RPM being 162/3 revolutions), it’d be best to average results from a number of trials. Repeatedly analyzing a recorded WAV file is a less tedious way of doing this than playing your test track multiple times.

I like the W&F results from this program because both quasi-peak and RMS numbers are produced from high sample rate data. The RMS numbers, calculated over 1s periods, while not exactly JIS (which specifies 5s periods, a 3000 Hz signal, and at least 30s of record), will in practice will be essentially the same as JIS so the results can be compared to Japanese spec sheets. And even if you’re not recording log files, after a few runs to get an idea of the variability of the results, it’s easy enough to stop the program at appropriate values of 10s quasi-peak and RMS to get representative results for a screenshot.

The Wow and Flutter Visualiser Audacity plugin is very cool and allows us to heed our guru @atomicbob ‘s oft-repeated counsel that looking at the waveform itself is more informative than looking at summary numbers derived from it. 3000 or 3150 Hz audio data (mono, or a single channel of a stereo dataset) are analyzed by the program and visualized as % variation about a centre frequency:



The data are presented raw with no just the 4 Hz weighting so the variability of the signal can be seen in all its high-resolution glory (44.1 kHz here as I used recorded WAV files). The ‘ideal’ signal would be a pure sine wave with 0.56 Hz frequency (10 cycles in 18 seconds, as displayed in this example), reflecting only the 331/3 RPM wobble of the test record, and if the record’s spindle hole was nearly perfectly centred the wave would have a small amplitude; but in real data, fluctuations on other scales are seen. In this case, there looks to be a spikiness of up to about +/- 0.05% at a frequency of around 10 Hz, and maybe some lower frequency cycles superimposed on that (more on this particular example later).

While the visualization is great, the program doesn’t actually give you standards-compliant W&F numbers or average rotation speed; you need to use one of the other apps for that. You can, however, visualize the data in other ways (again, more later) and if you were really obsessive you could download a text file of time vs % deviation (Sample Data Export from the Tools menu), do statistics on it to determine average speed and its variance, and analyze it with FFT to get the relative magnitudes of the different frequency components like this (you’d need to convert the % values back to actual frequencies first).

Edit: Another shortcoming of this method is that unlike with PlatterSpeed and WFGUI, the input audio data isn't filtered around the 3000/3150 Hz signal frequency, so any spurious environmental or electrical noise, common in the tens of Hz to 120 Hz - overlapping with the flutter frequencies - would be included. Applying some sort of filter to the audio data before analysis would be beneficial.

So, how best to use these different apps to compare different hardware; and, not least, how significant are any differences in results?

 

Using different apps to compare different belts

The tables below show the results of 3 measurements of each of my 3 belts using each of PlatterSpeed, Turntable Speed and WFGUI. Recordings for the PlatterSpeed and WFGUI measurements were made with the microphone about 1 ft in front of the tweeter of the R speaker, with the L speaker disconnected. In the WFGUI table, sets 2 & 3 are from different log runs of the same recording (same WAV file), set 1 is from a single log of a different recording. For the Turntable Speed measurements, the stylus (2 g of tracking force) was down on the outer track of the 180 g Tacet 12” record while the phone was sitting on top of a 45 RPM adapter placed over the spindle.



In terms of average speed, the PlatterSpeed and WFGUI results are almost identical, between 0.2% and 0.4% slower than the Turntable Speed results. It’s encouraging that the two apps that use frequency analysis are so close and not surprising they’d be fractionally different from the motion-sensing one. Turntable Speed suggests the Black belt is 0.2% faster than the Ref belt, whereas both PlatterSpeed and WFGUI give almost identical results for those two belts. Turntable Speed’s difference might be because of the smaller data records it uses - it apparently only analyzes 10s record lengths vs the nearly 30s used by other two apps in these examples. Also, it’s possible that the difference is coming close to the resolution of the phone’s motion sensor.

Comparing PlatterSpeed’s raw and notched fluctuations, though, with the W&F results from the other apps, PlatterSpeed looks like a huge fail: the PlatterSpeed results are all over the place, and some of the notch figures are larger than the corresponding raw ones. I’ve no idea how that should happen, and it’s not reflected on the graphic traces produced by the app. The Turntable Speed and WFGUI raw and W&F results are a lot more internally consistent, and more consistent when compared to each other. Together they suggest that the Blue belt performs worse than the Black and Ref belts, and that the two latter are not so different from each other in wow & flutter, as well as in absolute speed.

Presuming I don’t mind a small penalty in average speed (~0.8% below 331/3 RPM for Black & Ref vs < 0.2% below for Blue) in the pursuit of better speed stability, did I waste my money buying the new Reference belt instead of just swapping from my well-worn (2015) Blue back to my old (but actually pretty fresh) Black one? It’s possible that my results might be worse than ‘reality’ because I tested via a speaker rather than straight into the computer so there could be some environmental noise, and while my hub bearing, spindle and motor are only a few years old, my arm bearings date from 1984. Rega claims weighted peak (=quasi-peak) W&F for the Reference belt of < 0.08% vs 0.25% for an older one. Ignoring the PlatterSpeed numbers, all my results are inside that (suspiciously large!) range. To tell if they’re really different, it’d be useful to look at the waveforms to see how the speed behavior might differ in detail. So here are the Wow and Flutter Visualiser results, first for Blue, then Black, then Ref:

Blue


Black


Ref


These are clearly different. Ref is the ‘cleanest’ (most closely resembling a sine wave), has the smallest amplitude range between the sharpest peaks, and has the lowest amplitude variation across all the cycles; whereas on the others, the sharpest 0.56 Hz peaks are taller than their Ref counterparts, and many have chunks taken out of them by oscillations with frequencies of a few cycles per second – right in the W&F range. The Blue curve is by far the scrappiest, but even Black looks significantly worse than Ref, despite their similar W&F numbers.

The waveform differences can even be seen on the 10 Hz sampled PlatterSpeed datasets. Although the max raw fluctuations (yellow curves) are most extreme with the Ref belt (top), the Blue screenshot shows how other frequency components are superimposed on the 331/3 RPM swings:

Ref


Blue


The W&F standards all call for the maximum values of the various quantities to be the numbers that are reported. What looks to be going on here is that although the maxima might be similar in the different datasets, the average deviations are quite different; and that despite any frequency weighting, the 0.56 Hz (strongly record-shape related) oscillation will still dominate the results. So rather than having single maximum W&F numbers for each dataset, it’d be useful to be able to view them in terms of how much of each frequency is present across each. This is easily done in Audacity by selecting the ‘Spectrogram’ option from the track menu’s drop-down list.

The screenshots below have vertical axes displayed using the mysterious ‘Period’ option in the Spectrogram Settings, which is log-like but broader at the lower frequencies and has a minimum of > 2 Hz. Applying a maximum of 100 Hz focuses on the frequencies of interest. Purples are high values, greens (and white) are lows. Blue first, then Black, then Ref:

Blue


Black


Ref


The darker colours in the 3-6 Hz bins across the Blue and Black records vs the Ref record demonstrates higher contents of those frequencies in the former examples. So although the maximum W&F amplitudes might not be too different between Black and Ref, there’s clearly less total oscillation amplitude at those frequencies with Ref than with the other two belts.

The peaks around 10 Hz in all 3 displays are I think not due to turntable wow & flutter, but to the resonant frequency of my tonearm/cartridge assembly. Using the resonance test tracks on a couple of records, I’d previously concluded that max resonance of my setup is ~9 Hz.

I conclude from all this that although the ‘standard’ W&F specs with the different belts aren’t very different (particularly between Ref and Black), the actual low-frequency oscillation content of audio produced when using each belt will differ considerably. The important question is: to what extent might these differences be audible?

 

Listening (finally!)

The DIN standard calls for better than 0.2% maximum quasi-peak for ‘hi-fi’ turntables, implying that differences below that won’t be discerned by 'most' people. My Ref and Black belts gave essentially identical max quasi-peak (and 2σ) values and apparently run at almost identical speeds. The Blue belt is a bit worse in max W&F (although still under 0.2%) but runs a bit faster than the other two. Although being thoroughly polluted by expectation bias by this point, I spent some time listening for differences, particularly between Ref and Black.

And yes, I do think I hear some. The speed difference between Blue and the others, sure. It’s subtle, but hearing one straight after the other with the same track, it’s nonetheless evident. Differences that are likely due to wow & flutter I think are evident too: with the Ref belt, bass guitar and kick-drum attacks become tighter, to the extent that those instruments sound louder than with the others; and sustained string or keyboard notes (e.g. the cello in Tuxedomoon’s East and the organ in The Triffids’ Hell of a Summer) become less smeared and pitch changes between them more defined. Cymbal hits seem to become cleaner, too, which was something I wouldn’t have expected.

All these non-speed differences are more obvious between Ref and Blue than between Ref and Black, to the extent that even if there wasn’t a speed difference I might be able to distinguish Ref and Blue blind. Ref and black, hmm, doubtful; but even so I’m confident enough that differences between those two are audible, even with my ears on my equipment (the rest of which, for completeness, was Denon DL-110 cart, ifi Zen Phono preamp, MCTH amp and HD6XX headphones).

So what have I learned from all this, and what questions remain?

• Wow & flutter specs aren’t straightforward to compare, especially when obtained in different ways
• A “good enough” wow & flutter result doesn’t necessarily tell the whole story. It’s more useful to look at the waveforms or spectral properties of the low frequency content (familiar refrain). This is pretty easy if you have a decent test record (critical!), a laptop to plug your turntable into (less chance of pollution than making a recording) and Audacity
• The Rega ‘Reference’ belt has better W&F performance (as distinct from measured specifications; or rather, specifications derived from my measurements) than my two other belts. My other two are older and my results might be unfair to the Blue belt in particular, because that one’s had heavy use. The differences were audible to my ears on my equipment
• Why are my measured differences so distinct from the manufacturer’s (Rega link above)? Sure, I wouldn’t expect my measurements of the new belt to be as good, but I’m surprised my measurements of my old belts are apparently so much better than Rega’s measurement of a (presumably) fresh old belt. Did they use a really old old belt, or a different test record? The 331/3 RPM swings in their ‘old’ example are extreme
• To what extent does the Reference belt run slow on all Rega pulleys and sub-platters? It seems to be common experience (see the reviews here) but can be compensated for with the adjustable Rega Neo PSU. Would belt tightness (it fits looser than the others) make a difference to speed, without adversely affecting wow & flutter? If so, moving the motor mounting slightly (I could DIY that on mine) might be a way to bring speed closer to spec; I can't adjust speed on my older PSU
• Finally, is the Reference belt worth the money? You decide! Because Rega has sole-distributor agreements for markets outside the UK, UK retailers typically won’t ship one to you and you’ll pay a significant markup over the UK price to buy locally. For example, in Canada the $95 before-tax price is over twice the UK before-tax price, and in the US the $75 price is 2.5 x. I bought mine in the UK and paid the 20% VAT (28 GBP total).

 

My guess is that this belt is great for older/standard models and I don't recall but suspect that units with outboard sinewave generators may have internal pots to be able to correct for this belt. I suspect speed difference is due to this belt perhaps being fractional smaller in diameter and runs closer to the spindle since the motor pulley is a tapered V.

 

The new belt is definitely slightly thinner than the Edwards blue (although less than the resolution of the vernier on my crappy caliper), which in conjunction with the V profile of the stock pulley might indeed explain the slower speed. The Neo PSU is speed-adjustable, but the older TTPSU, which I have, sadly isn't.

I haven't yet played around with the motor mount to see if increasing the tension might raise the speed, but I think it'd be more likely to do the opposite because of exactly what you say: the belt would ride further inwards on the V and reduce the pulley's effective diameter. If the low belt tension was giving rise to significant slipping, I'd imagine the W&F would be worse than it is.

Still might try it though, if only for science

 

Speaking as a guy with negligible turntable experience, and absolutely zero with belt driven... is there anything that's used to increase friction of the belt? (like rosin or something)

 

The usual suggestion is to keep it and the pulleys as clean as possible. Putting any compound on it to increase friction might risk making it stick to one or both pulleys, which would be counterproductive.

'Stick' meaning, hold on beyond the tangential point of the path to the other pulley.

 

Some thoughts again coming at it from a purely outsider perspective...
- increased friction on flat pulleys is fine unless the surface has an actual stick to it
- with a v-belt, overall effective friction is higher due to the higher perpendicular surface areas and perpendicular forces on the belt, which requires enough much trigonometry for the typical high schooler to run away screaming
- but also with a v-belt if the tension actually pulls the belt tighter into the groove (assuming just a single V, not those belt with the multiple v-grooves so essentially just a flat belt ribbed for turntable pleasure) then the compression will generate a self-stick to release which I imagine is undesirable

 

These belts are circular in section. It's the motor pulley that has a < profile so with increased tension and consequent reduction of belt diameter, the belt will go deeper into it - leading to reduced effective diameter of the pulley as well as possible self-stick. Extra friction on the motor pulley would I'd think increase the tendency for self-stick. Extra friction on the flat-faced driven pulley (sub-platter), though, should be fine.

 

Do people experiment with different size belt cross-sections? Otherwise I'd think a semicircular or even elliptical groove would be better than a V and would reduce the sloped stick thing.

 

Probably. Alternate belts are available from a few manufacturers other than Edwards, but I haven't been down all the Rega online rabbit-holes to find if people deliberately try different sizes, or fishing line etc. as they do with some other tables.

I'd have thought so too. Apparently the designer "has always championed the use of a circular drive belt and a 'V' shaped pulley", but then he's always been an iconoclast. A pulley I have from another manufacturer is also V-shaped, and looking online I see a couple of others the same. Perhaps the argument is that the inside of the belt might be rougher because of the molding process, whereas the belt's width measured across the chord of the contacts with the V-shaped pulley walls is expected to be more uniform?

 

The V is more universal in that regardless of the diameter of the belt cross-section, the contact surface and math works out mostly the same. If you have mismatched circular grooves and belt sections, the math is much harder... but within a certain range of tolerances your contact surface is higher. Hmm but maybe the amount of friction doesn't pan out... I'd have to see the math on it.

 

Some more thinking out loud... if there are any variances in the belt elasticity along the length of it, increasing tension could potentially even it out since the "soft" portions would stretch first until it equalized with the rest of the belt.

Really I'm just intrigued by this from the material science and stress analysis point of view.

 

But if the elasticity variations correlated with thickness variations with the thinner parts being the more stretchy, increased tension would exaggerate the thickness differences and thereby exaggerate variations in effective pulley diameter.

The Korf Audio blog has a couple of posts on belt drive considerations that might interest you. I'm not sure I buy everything he says given some of his quantitative results in earlier posts, though.

 

Very unlikely that it's slipping when at speed. Very little energy/tension needed to maintain same. For those with Regas. give the platter a push with your finger at startup and your belt will last a very long time. About the only things that causes wear is startup or a very long period of time.

Also, I haven't liked any mod I've experienced with them. They are well voiced as is and different isn't better.

 

If there are variances in the thickness, then yes the thinner part would stretch first because the stresses are higher there which would lead to inconsistent belt loading as it whips around.

But it's also possible that with consistent thickness that the material properties might vary as well. In this case, the "softer" sections would stretch first and you'd have the same problem.

I'm guessing this is one reason why some systems will have multi-belts, to smooth out the output (hmm, a little like paralleled dac chip outputs).