Dynamic Comparison of CD, DVD-A, SACD - Part 2

My previous comparison of 2-channel recordings of the title track of Diana Krall's The Look Of Love on CD, DVD-A (Group 2 - MLP 2.0 96/24) and SACD (2-channel) versions yielded a surprise in terms of the relative dynamics: the CD version was noticeably more "compressed" in dynamics than the DVD-A version, which in turn was also compressed relative to SACD.

I have yet to read any literature on the respective formats which would hint at this result that I began to doubt whether this might be an anomaly of the way I did the three recordings. I did use different gains for each recording, and I wondered whether a different gain setting may have contributed to altering the dynamics of the three recordings.

Therefore, I resolved to re-do all three recordings using constant gain, to try and see if I can replicate the dynamic differences for the three recordings.

This time, I've also done the recordings more accurately. I've abandoned my old copy of Cool Edit as a recording tool, although I am still using it to generate statistics and graphs. I discovered two things:

• Cool Edit does not seem to support 24-bit recording on the Prodigy 7.1 - I discovered using Cool Edit's "Statistics" function that I was actually recording at 16-bits!
• Cool Edit only supports the WDM/MME driver, which means the samples are probably passing through Windows' kmixer (also known as "kmangler" to audio enthusiasts due to it's propensity to resample audio streams when you least expect it to)

I am now using FASOFT's n-Track Studio Lite which comes bundled with the Prodigy 7.1. n-Track supports Prodigy's lower-latency and more transparent ASIO driver instead of the default WDM/MME. I've verified that n-Track does support 24-bit recording. I am also recording "silent" (with no monitoring of the signal) to prevent any possibility of cross-talk between the playback audio circuits and the recording audio circuits.

Well, here are the results ...

Statistics
Recording the three versions at constant gain clearly shows that the SACD recording is around 3-4dB lower in level compared to the CD and DVD-A recordings. In an A-B comparison between SACD and CD or DVD-A (on my system), a typical listener will have a tendency to pick either CD or DVD-A as "better sounding" if the relative levels are not adjusted. But as we shall see, once the relative levels are adjusted, the increased dynamics of the SACD recording are confirmed.

This is what Cool Edit says about the relative statistics of the three recordings (using an RMS Window of 50ms where 0db=FS Square Wave):

CD

DVD-A

DSD (unnormalized)

DSD (normalized)

Given that the level of the DSD recording is so low, I decided to "normalize" the DSD recording:

In Summary

Comparing the average values for all four:

Note that by looking at the difference between the Maximum and Average RMS Power (second last row), the statistics support the hypothesis that DSD has slightly higher dynamics (11.66dB) than MLP (11.55dB), which in turn is slightly higher than CD (11.47). Note also that normalizing the DSD recording does not change the relative differences.

You may think the difference may not seem significant (only 0.11dB), but remember we are looking at differences between the Maximum and "Average" RMS Power, therefore the differences (which are more prominent the louder the signal is) have been averaged out.

The DSD difference between Maximum and Minimum RMS Power (last row) is relatively low at 76.41dB compared to MLP (82.45dB). This can be explained by the average level of the DSD recording being less than the MLP recording (the underlying A/D accuracy of the Prodigy sound card may be limiting the accuracy of the lower level DSD recording) as well as the DSD ultrasonic noise floor.