Understanding Distortion

In the previous article in this series I gave an outline of the behaviour of LP replay systems. Now I can introduce a new application, !TrackTHD, which you can use to assess the performance of your HiFi system. Here I’ll concentrate on the effects of playing weight and bias adjustments on LP replay. But you can, of course, use the application to determine the distortion level from other sources.

Installing and using !TrackTHD is similar to the previous applications I’ve introduced in this series, so I won’t repeat all the details again. If not sure how to get it going, please refer to the previous issues of Archive and/or see the !Help file in the application.

When you run !TrackTHD it opens a taskwindow and asks you for the name of the sound data file to analyse. It then behaves like !TrackTrans and asks for a starting time for the section to examine. Once you’ve given this time (in seconds from the start of the recording in the sound file) it loads a section of sound data 50 milliseconds long, starting at the instant you specified.

The application then prints various items of information in the taskwindow, and creates four files on your ramdisc. The details of the contents of these files are given in the !Help file, and they can be used to produce displays of the waveforms and spectra. To illustrate this have a look at Figures 1a and 1b.

These show two graphs which I produced using !Tau to plot the contents of two of the files saved to the ramdisc. In each case the red line shows the left channel of the stereo, and the blue line the right channel. I’ve just shown a 10 millisecond portion in each plot for the sake of clarity.

fig1a.gif - 21Kb
Figure 1a

Figure 1a shows the contents of the file “waves_<name>” created by !TrackTHD where <name> was the name of the input file. This displays the waveforms. In this example the test waveform I recorded was nominally a 300 Hz monophonic sinusoid from a test LP.

fig1b.gif - 22Kb
Figure 1b

Figure 1b is a plot of the contents of the file “nulled_waves”. This shows what the pattern looks like if you null out (remove) the nominal sinusoid to reveal any distortion and noise. The null-waves pattern is the kind of display someone using a professional distortion analyser would see and is a useful visual guide showing the signal details that, ideally, would be absent!

In order to obtain this nulled result the application has to determine the frequency, amplitude, and phase of the intended sinusoid. Alas, although the test LP may say this is “300Hz” this may not be the case for various reasons – e.g. the turntable was rotating at slightly the wrong speed. The application could try to determine the frequency by using Fourier Transformation. Unfortunately this does not work very well for short duration signals like our 50 millisecond section. But a longer signal to analyse exposes us to the problem that the actual frequency may wander around due to the turntable not rotating at a perfectly steady rate, or the LP being pressed slightly off-center.

Given a periodic waveform whose amplitude is much greater than the noise level we can use a ‘zero crossing’ or ‘fringe counting’ method to determine the signal frequency. The way this works is illustrated in Fig2.

fig2.gif - 17Kb
Figure 2

We can scan through the sample values looking for pairs of successive points like ‘A’ and ‘B’ where the one value is negative and the next is positive. These pairs surround a zero crossing of the periodic waveform which we can regard as the start of a cycle. If you look at the output from !TrackTHD in its taskwindow it reports what it finds for each channel. For the example I used for the results in this article this reported that “Left 126 to 2163 has 14 cycles”. This means that the first such crossing occurred between samples 126 and 127, the last between 2163 and 2164, and that there were 14 complete cycles in between these crossings. Knowing that there are 44100 samples/sec this allows the application to work out a value for the frequency of 303.09 Hz in this case.

For short-duration, high signal/noise, data that represents simple periodic waveforms this method is much faster and more accurate than the conventional Fourier Transform spectrum based approach. In this case, the frequency can be determined with an accuracy of better than 0·1%. The application carries out the above for both channels, and then selects the result from the loudest channel. It then uses this as the fundamental frequency for the THD analysis and spectra.

The application now works out the spectra of the Left and Right channels, and saves the results in the “raw_spectrum” file on your ramdisc. The resolution and range of the spectra are controlled by the value of the fundamental frequency found by the fringe counting. It then prints out in the taskwindow the results for the fundamental frequency and the first few harmonics (integer multiples). Once this is done, the application proceeds to null the fundamental component from the Left and Right waveforms, and repeats the process of saving the resulting spectra and printing out the harmonic levels and THD in the taskwindow. The nulled spectrum is saved in the “nulled_spectra” file on your ramdisc.

Note that the spectral results saved in the files, and displayed in the taskwindow show three sets of results – Left, Right, and Total. The “Total” value is simply the power sum of Left and Right. This value is the relevant one for monophonic (horizontal) or difference (vertical) modulation, but the Left and Right values are useful for diagnostic purposes.

fig3.gif - 18Kb
Figure 3

Figure 3 shows the spectra I obtained for the example I have used above. Here I have plotted just the “Total” spectrum for both the raw (broken line) and nulled (continuous line) waveforms so they can be compared. The broken line demonstrates one of the limitations we would face if we simply Fourier Transformed the input data without nulling. The relatively large component at the fundamental frequency (303Hz) produces ‘wings’ in the spectrum that tend to hide the presence of the harmonics at much lower levels. By nulling we can reduce these unwanted wings to reveal the actual distortion components. Thus when using !TrackTHD it is generally sensible to use the results for the THD, etc, produced from the nulled waveforms.

Having outlined how the application works, and the results it produces we can now consider how to make use of it, and interpret the results. In this case the signal modulated onto the LP was nominally a 300Hz mono sinewave and the LP sleeve stated this has a level of “+14dB relative to 1·12 x 10-3 cm peak amplitude”. This immediately runs us into the problem that although the industry standard reference level for LP is specified as a velocity of 5cm/sec at 1kHz. we now have a signal at a different frequency specified as an amplitude! Thus we need some way to convert between these to help decide if this signal is soft, loud, or very loud. Fortunately, I already did the necessary calculations and gave some results in the previous article as I knew this problem was looming. I’m afraid that it is common for test LPs to specify sound levels in various ways – often in different ways on the same LP! If you have the last article to hand you can look at the graphs I showed there. However to make the situation clearer we can use the following table of results for a few commonly-used frequencies.

RIAA 0dB Reference Level values

(Hz) Gain (dB) (cm/s) (m) (g)
20 19.21 0.55 43.60 0.070
100 13.03 1.12 17.75 0.715
300 5.45 2.67 14.17 5.135
400 3.75 3.25 12.91 8.318
1k 0.00 5.00 7.96 32.037
3k -4.73 8.62 4.57 165.691
5k -8.20 12.85 4.09 411.738
10k -13.72 24.28 3.86 1555.599
20k -19.61 47.81 3.80 6126.766

We can now use the table to see that at 300Hz a stylus velocity of 2·67 cm/sec would mean an amplitude of 14·17 microns. This allows us to translate the value shown on the LP cover and it turns out that the text waveform in this case has a level of +12dB relative to the RIAA reference level - i.e. fairly loud. The THD levels worked out by !TrackTHD were:

Left -32.24 dB ( 2.44 %) Right -40.09 dB ( 0.98 %) Total -34.31 dB ( 1.92 %)

if you are not familiar with the distortion levels an LP replay system tends to produce then the above values may seem high. However they are in fact quite low for such a loud signal, and just under 2% for the total (mono in this case) is quite a good result for a +12dB sinewave. The result therefore immediately gives some indication that the stylus is OK. However to be more confident of this we’d need to check the value for a number of test waveforms, etc, to ensure the result was not a lucky one.

The next thing to note is that the Left channel value is over double that for the Right channel. This indicates a definite asymmetry in the performance. Looking back at the nulled waveforms in Fig1 this is confirmed as we can see that the Left channel residual is much larger than the Right channel residual. The signal levels for the fundamental also differ slightly with the Left channel being 0·2dB louder than the Right channel.

Again, for a sensible diagnosis we’d need to repeat the test with various signals before coming to a conclusion. However here let’s assume these differences persist. We can now experiment with slightly altering the playing weight (downforce) and bias force to see what effect this has. If you do this you will probably find that there is a tendency as the bias (sometimes called “anti-skating”) force is adjusted for the relative distortion levels to change, with the THD on one channel rising as the other reduced. Changing the playing weight tends to alter both channels together. Hence if you get a result like the above you may find that altering the bias adjustment can reduce the distortion of the ‘poorer’ channel – possibly at the expense of making the ‘better’ channel worse. So if you have the patience you can experiment with adjusting the playing weight and bias settings to see if you can reduce the overall level of distortion, checking as you go by making another recording and analysing the results with !TrackTHD.

A few words of caution, however. Take care not to adjust the playing weight too far outside the range the manufacturer specifies as this may lead to damage. It is common for the distortion to be lower with a playing weight slightly higher than the maker recommends. However this may lead to excess record wear, or stylus failure in some cases. When playing the test bands, tend to start with the lower signal level ones, and listen carefully for any ‘buzzing’ effect on the sinewave. This is a sign of mistracking or gross distortion. Note quickly if this occurs on one channel more than the other. If so, the bias may need adjusting. If it occurs on both channels then the stylus is having trouble playing the disc and the playing weight may need increasing if it isn’t already as high as it will safely go!

Some of the distortion you encounter may be recorded (or worn!) onto the test LP, so the results may be misleading unless you have new LP of excellent provenance. Finally, bear in mind that the internal arrangements of the cartridge may be slightly out of alignment, or the arm may itself need re-alignment. In this article I’m assuming the arm and cartridge geometry and setup correctly, and the turntable is horizontal so I’ve not discussed testing and adjusting these. So a difference between channels may be due to factors which adjusting the weight or bias don’t really correct. Although if people wish, and Paul is happy for me to do so, I could explain that in a later article. All being well, I hope in later articles to explain how you can use various members of the !Track application family and your RO computer to check and adjust other aspects of your HiFi.

Jim Lesurf
1st Mar 2007

returnblue.gif - 6Kb