Dynamic Comparison of CD, DVD-A, SACD - Part 2 - page 2
Peak value analysis
The peak sample for the normalized SACD recording occurs at 3:42.475 seconds into the song (or sample #21,357,637!).
At this point in time, the normalized SACD waveform looks like this:
In comparison, the CD waveform is definitely more compressed, and the three middle peaks on the right channel show the tell-tale "flattening" from clipping:
Interestingly enough, even the DVD-A waveform exhibit clipping (but far less prominent):
Comparing the SACD and DVD-A recordings side by side clearly show that SACD is relatively more "dynamic" than DVD-A:
DVD-Audio (left); SACD (right)
In the above waveforms, the DVD-A recording has been attenuated by -1dB so that the relatively quiet segment at the beginning of this extract are approximately the same amplitude for both DVD-A and SACD. In contrast, the amplitudes of the the middle, louder, segment, is at least 0.5-1dB louder on SACD than on DVD-A, thus demonstrating that SACD has higher relative dynamics.
Histogram analysis
Now comes another surprise! But before I explain, let's look at some histogram graphs of the distribution of levels.
First of all, this is the histogram of the CD recording:
This shows that the actual dynamic range of the recording is only around 30dB (from -35dB to -5dB). There is negligible musical information below -35dB. Note the "hump" at the extreme left which corresponds to the "hump" around 20kHz in the spectral analysis.
The DVD-A recording looks very similar:
Note that the shape of the histogram is "narrower" in width around -35dB to -5dB. This indicates that the DVD-A recording is of a higher resolution than CD, as more values are spread across the entire histogram instead of concentrated around the median.
By comparison, the "normalized" SACD histogram has a strange "quirk" (the unnormalized histogram is similar):
Notice that there are no samples between -78dB and -62dB!!! In contrast, there is a higher proportion of samples between -62dB to -35dB, and the shape of the histogram is also "fatter" in width between -35dB to -5dB.
This is surprising indeed, because it implies that there are no samples below -62dB!
It turns out that this is caused by our old friend: DSD ultrasonic noise. Applying a 6-th order Bessel low pass filter at 20kHz - which looks like this:
... provided a "normal" looking histogram.
Since then, I have recorded other titles and formats and discovered that this phenomenon of "missing samples" below a certain threshold is quite common and tends to occur when the original source is noisy. I have encountered this in recordings from LP as well as DVD-As of analog sources.
Conclusions
Well, I was hoping merely to confirm that the DSD recording has higher dynamics than the PCM recordings on CD and DVD-A, and I did manage to confirm my previous observation. The differences between the dynamics can be due to many reasons, the most likely being differences between the players in their ability to reproduce dynamic transients accurately.
However, I did not expect to see clipping on the CD and DVD-A recordings. This could either be a fault in the transfer, or in to D/A conversion stage, but it's quite interesting that I was able to observe clipping on two separate players (the Sony playing CD and the Panasonic playing DVD-A)! This clipping may be the reason why many people feel PCM recordings sound "edgier" than LP or SACD if it is prevalent during playback!
© 2004 Christine Tham
Reprinted with the Permission of Christine Tham. Please visit her very informative website at: http://users.bigpond.net.au/christie/
