A while ago Hi Fi News published an article where I compared the dynamics of BBC Radio 3 on FM, Freeview, and DAB[1]. This showed that FM did, indeed, differ quite significantly from Freeview and DAB. I then started wondering if an LP/CD comparison might also show differences. The results I found were quite shocking.
I started with a piece which I have in many versions, spanning a number of years. Although I mainly listen to classical music and jazz these days, I also exhibit some signs of the aging 60s drop-out. One feature of this is a liking for Jimi Hendrix. I therefore chose to use versions of ‘Purple Haze’ to examine in detail. I have more versions of this than I’d like to admit, but Figure 1 shows the results of my analysis for just four of them. As in the previous article, these plots show what percentage of the time the sound is peaking at a given level.
If you look carefully at Figure 1 you can see differences between the 1978 LP and the CDs from 1985 and 1989. The CDs show a steepening of the ‘leading edge’ of the dynamic distributions which implies that they have been level compressed compared with the LP version. This is a sign that they have been altered in ways that may be audible. However, around the year 2000 there was a major reissue of Jimi’s work under the “Experience Hendrix” logo. If you compare this version with the others the differences are quite dramatic!
With CD audio, no recorded sample value can exceed the 0dBFS level. Yet this version of Purple Haze spends over 30 percent of the time within 1dB of this maximum possible level. No wonder that it sounded different to previous versions.
I then investigated a selection of CDs and LPs of various vintages.
It became clear that it was quite common for recent CD versions to have dynamics different to those on older LPs. Figure 2 is an example. A 1994 CD version of ‘ShaLaLaLee’ by the Small faces has dynamics which are obviously level compressed compared with a 1984 LP version. The irony here is that although CD has the ability to provide a wide dynamic range, such pop/rock CDs make no real use of this. In simple engineering terms, there is no need whatsoever to level compress the recordings on CD in this way. Yet I found that it is surprisingly common.
Things got even more worrying as I investigated the possibility that some versions were clipped. To do this, I wrote a new program. This scans the sound data looking for sequences of identical (or very similar) successive sample values close to the maximum values possible for CD. The results allowed me to find potential examples of clipping, and see how often they occur in a given CD track. Figure 3 shows a couple of examples. One is a section of ‘Purple Haze’ from the ‘Experience Hendrix’ CD version I mentioned earlier. The other is from a sampler CD and is Horace Silver playing “The Hippest Cat in Hollywood”. (Impulse! label, IMPD-207, released 1996). This track has one or more successive samples within 0·015dB of 0dBFS on over 1,000 locations throughout the track. The ‘Purple Haze’ example has over 2,500 such events! Note that for clarity I have scaled the plots so that +/-1 represents the maximum level possible for CD. With due political correctness in mind, I use a red line for the left channel, and a blue one for the right.
It is important to bear in mind that I am not singling out these examples on the basis that they are unique. I found similar effects on various CDs. Classical examples were harder to find but I found a few, one of which was particularly surprising, illustrated in Figure 4.
This shows both the dynamics and an example of clipping from a recording of the 1812 Overture. The track is from a “Mercury Living Presence” CD, 434 360-2, released 1992. According to the results I obtained, this track contains well over 9,000 events where one or more successive samples are within 0·015dB of the maximum level possible for an audio CD! As a Living Presence recording, the original would have been made using analog tape, etc. Hence the ‘flat top’ clipping shown in figure 4 seems unlikely to have arisen in the original, although alas I don’t have an early LP of this, so can’t compare one with the CD. But it would not surprise me if the LP sounded very different to the CD.
In some cases the clipping seems to be ‘soft’ rather than flat-top in shape. Figure 5 shows an example of this taken from a track, “Ain't No Sunshine”, by Al Jarreau on a freebie CD given away with a newspaper. The histogram of peak powers shows clear signs of level clipping and/or excessive compression. But the waveforms show that as the level approches the limits it tends to curve softly towards them rather than having sharp flats. Hence although the results in some cases seem indistinguishable from the actual CD waveform being allowed to clip, in other cases the effect may have arisen in an earlier stage during the recording or production process.
In other cases clipping or saturation seems to show up, but may occur at a level well below the maximum possible on CD. An odd example of this is shown in Figure 6. This was taken from another freebie CD given away with a newspaper. In this case the track is by Franz Ferdinand. Looking at the histogram it clearly shows some sort of clipping or limiting. But this seems to be limited at the -3dBFS level of the CD. The result looks like clipping was applied an some stage prior to the data being written to the CD.
As I investigated this topic I also spoke to someone I know who is an experienced professional in the recording industry. He mainly works on classical and jazz recordings, but also has long experience and attends pop CD mastering sessions on a regular basis. He commented that good sounding, well produced studio masters were sometimes subject to futher compression, etc, to create “maximum impact”. This was regarded as an important criterion in pop recording. The argument went that, people didn’t care about clipping or compression and that “louder” was “better”. He also admitted that he sometimes came away from such sessions shaking his head at what they did. But as a professional, his job was to satisfy those who employed him.
Following these conversations I found that the topic of the “Loudness War” has been commented on in the past on the web. However I don’t think that most audio enthusiasts or music lovers have been aware of this, or the implications. I certainly have not until now, but this may be because I mostly listen to classical music or jazz, and most of my rock/pop recordings are old LPs. So this passed me by!
Does all this matter? Certainly as a classical/jazz listener I don’t want the recordings I play to have been clipped, and I am wary of level compression. When I decide to replace an old LP with a newer CD re-issue of some old rock/pop music, I don’t want the CD to sound different as a result of being compressed or clipped. I also wonder if this is always deliberate, and how often it may occur by accident, or unawares.
Is this one of the reasons people say that the ‘dynamics’ on LPs are ‘better’ than on CD? Are the recording companies creating a situation where we often may be unable to compare LP with CD on a fair basis? It seems bitterly ironic that the CD medium offers a dynamic range of over 90dB, yet what is recorded may be compressed and clipped into a far smaller range!
Jim Lesurf
29th Sep 2006
[1] Sound on Vision Hi Fi News pages 70-73 March 2006
