Ultrasonic noise in some high resolution downloads

[scan80269]

I'm probably late to the party, but I've been discovering significant non-musical ultrasonic noise embedded into certain high resolution PCM music downloads.

 

For example, the 2L.no test bench site carries evaluation samples of recordings at various resolutions & formats.  I ran MusicScope on three files for the 2L-106 Arnesen Magnificat album:

 

DXD (2L-106_stereo_DXD-352k_MAGNIFICAT_04.flac)

24/192K (2L-106_stereo_PCM-192k_MAGNIFICAT_04.flac)

24/96K (2L-106_stereo_PCM-96K_MAGNIFICAT_04.flac)

 

Attached screen shots are with MusicScope stopped at the last second of the tracks, where the music has faded to quiet.

 

Among these three, the DXD file has the largest hump of persistent ultrasonic noise, going up to around -60dBfs at 176KHz.  The 24/192K file has a smaller hump peaking at 86KHz.  Only the 24/96K file appears devoid of any significant non-musical ultrasonic noise.

 

Any idea as to why a high resolution PCM file would contain such ultrasonic noise?  The noise humps remind me somewhat of DSD, but can they be the result of noise shaping?  If the master recordings were DXD (24/352.8K), it seems to me the dynamic range of the container is more than big enough to capture all the music, and noise shaping or dithering should not be needed.

 

I'm a bit concerned as to how such ultrasonic noise can affect a playback system.  For example, my Auralic Vega DAC performs internal upsampling of incoming PCM to 1.5MHz and uses "gentle" low-pass filters at its outputs.  With a high resolution PCM recording carrying significant ultrasonic noise (like the DXD file above), these's a risk a certain amount of ultrasonics will leak out of the DAC to reach downstream analog equipment like preamp or headphone amplifier.  My Auralic Taurus Mk II headphone amp has a frequency response up to 300KHz (-3dB) so it is capable of being affected by ultrasonics sent from the DAC.  The ultrasonic noise being way above 20KHz is not directly audible, but the intermodulation effects of this noise with the music in the audio spectrum (within 20KHz) is potentially audible to some extent.

 

I have discovered for some time that inserting a Jensen ISO-MAX PI-2XX XLR balanced isolation transformer between my DAC and headphone amp can reduce the harshness & brittleness of treble when playing some high-res PCM files.  The transformer acts as a low-pass filter to attenuate the amount of ultrasonics reaching the headphone amp.  This finding leads me to suspect any ultrasonics embedded in high resolution PCM recordings may not be totally harmless.  In other words, the ultrasonics in these files may be preventing them from sounding their best in my system.

 

As a separate ongoing experiment, I've been running XiVero's XiFEO FLAC entropy reducer on a number of 2L sample recordings (w/ DXD master) that have the ultrasonic humps.  XiFEO can reduce encoded noise below the noise floor (by LSB truncations & flat dithering) and apply LPF to band-limit the content to achieve higher FLAC compression ratio.  In some cases the FLAC file output by XiFEO is <20% the size of the input FLAC!  A quick A/B listening test at a friend's place resulted in the processed FLAC being declared better sounding than the original FLAC.  The tested system benefits from the offline removal of embedded ultrasonics in such files.  Once XiFEO is enhanced to support manual setting for the LPF cut-off frequency, I plan to re-process my ultrasonic-carrying high-res PCM files followed by some serious listening to compare with the original downloads.

 

[scan80269]

1 hour ago, mansr said:

Virtually all ADCs use a sigma-delta (DSD-ish) front-end followed by a digital low-pass filter. At 176/192 kHz and higher output rates, some modulator noise is usually present at the top end, and this is what you're seeing here.

 

This is a valid concern. The obsession with gentle filters combined with the inevitable high-frequency noise often does more harm than good.

 

The main reason for using high sample rates is to ensure any filter artifacts are well removed from the audible range, and 96 kHz is plenty for this purpose. Since most microphones only extend to 50 kHz or so, and even that is uncommon, higher frequencies in a recording have little but unwanted noise.

 

My advice, save your money and buy the 96 kHz version.

Thanks for the clarification, mansr!

 

I also found some strange behavior regarding ultrasonics in the first track of this 2L-106 Arnesen album in DXD.  While observing the playback of this track with MusicScope, I noticed the ultrasonic noise dropping off abruptly at the 5:25 mark, then returns for a fraction of a second at the very end of the track.  It looks as if someone played around with LPF controls during the mastering of this track.

 

I suppose doing the master recording at DXD is also intended to support conversion to DSD for release?

 

Your advice is excellent indeed and I will heed it.  Even Barry Diament's fantastic recordings released at 24/192K show very little musical energy above 30kHz or so.  For a handful of 2L recordings I bought previously at DXD resolution, I will do offline LPF to get rid of the ultrasonic noise.