MiniDisc Generational Loss Tests
One weakness MiniDisc has when compared with DAT is that perfect
copies cannot be made (without professional equipment) due to the
asymmetric nature of the MiniDisc audio compression system. Every time
the signal goes through the compression/decompression process, it
changes slightly. After several generations these effects become
noticable.
One could argue that whether or not there is loss from generation to
generation is a moot point since SCMS (Serial Copy Management System
-- present on all consumer digital audio equipment) allows only a
single generation of digital copying.
However professional machines exist which do not obey SCMS, and so the
question remains, how bad are the effects of this generational loss?
In order to discover this, I conducted an experiment which produced
samples of audio copied up to 100 times.
I generated these samples as follows:
- Digitally copied from CD (Panasonic MASH SL-S490 portable player)
to MiniDisc (Sony MDS-503 MD Deck), a short audio sample (Donald
Fagen: Kamakiriad, Trans-island Skyway (track 1), 1:34-1:54)
- Digitally copied this segment back and forth from the MDS-503 to a
Sparcstation 2 via a Townshend Computer Tools "DATlink" (workstation
SCSI digital audio interface) 100 times (over the optical connector),
saving each intermediate copy.
SCMS was not operative during these copy operations since the
professional equipment used was not set to generate a copy restricting
SCMS code.
A selection of the resulting generational copies are given below.
(You may need to equip your
Unix,
Macintosh,
or
Windows
system with an MPEG audio player to hear
these.)
All samples are 20 seconds of stereo audio, 952KBytes each, MPEG Layer
II at 384Kbit/sec.
I have chosen not to do subjective evaluations of these samples for
the following reasons:
- It is a very serious undertaking. I have talked to several
researchers here at ATR about
the science of subjective testing, and if one wishes to perform a
statistically valid test, it requires a good deal of work.
- The effects of compression vary greatly depending upon the source
material. Tests conducted on one musical sample will not hold for
others. Selecting the samples themselves is a difficult task.
- The few subjects I have tested have slowly become attuned to
compression artifacts, causing a sample that was originally judged to
be transparent to show artifacts on subsequent presentations. For me
this calls into doubt the entire premise of a subjective evaluation of
MiniDisc generational loss, and prompts the question of what, precisely,
I would like it to prove.
Based upon the reasons given above, I have chosen to simply provide
these samples so that others may judge for themselves the effects of
MiniDisc generational loss.
Eric Woudenberg, April 1996
Addendum
The reasons for generational loss are examined in Frank Kurth's
paper An Audio Codec for Multiple
Generations Compression without Loss of Perceptual Quality (pdf).
