Home Page | Up
One Level | Site Map | Send
Us email
ADPCM vs. MPEG Audio
5/13/96
In attempts to make some progress on the MPEG vs. ADPCM encoding issues,
a series of tests were conducted, with the goal of eliminating or controlling
as many variables as possible. The original complaints were with CD-i
developers using MediaMogul versions 2.2 and earlier. Current tests were
conducted with MediaMogul 2.2.1 and corresponding plug-ins. A similar
disc using Balboa was also produced to eliminate MediaMogul as a variable.
Test Conditions:
- The source audio was a digitally generated 1 KHz sinewave, monophonic,
44.1 KHz sample rate, 30 seconds in length, in AIFF format. The amplitude
swing was +/- 50%. This produced an -6 dB signal level relative to full
16 bit modulation--to eliminate amplitude clipping in ADPCM or MPEG processing.
This -6 dB absolute level was normalized, subsequently termed "0
dB" and used as reference throughout the tests.
- The AIFF source was down sampled to 37.8 KHz using Sound Designer
DSP tools. The amplitude was verified to match the source file. I used
AudioStack on the Macintosh (the same tool a developer would use) to encode
the resampled file to A level, Mono, ADPCM.
- The AIFF source was also encoded into 192 kbits/sec, Mono, MPEG audio,
using Audio2MPEG on the Macintosh. The intent was to remove hardware A/D
converter and hardware encoding issues. The elementary audio file was multiplexed
with a nondescript elementary video file, using PINK. The multiplexed file
was processed into an RTF using Master. This stream is designated MPEG
#1.
- Another MPEG stream was prepared by encoding a Beta SP tape of color
bars and 1 KHz tone. The A/D converter gain at the Delta Vx encoding system
was adjusted so that the "clip" light just flickered (-1.5 dB
from full 16 bit modulation according Apogee documentation). This is the
only "metering" provision on the encoding system. The audio was
recorded at 192 Kbits/sec, multiplexed using PINK and RTF'ed by Master.
This stream is designated MPEG #2.
- Both the ADPCM and MPEG assets were scripted in MediaMogul 2.2.1 using
DV Plugin 2.2.1 and processed with Script2Disc 2.2.1. The default audio
attenuation settings were used (no VOLUME or $ command was used to manipulate
volume of either asset type).
- A CD-i disc of the build was recorded using CD-IT All 2.5.8. The
disc is marked: Level Test (Audio) 5-15-96 MM 2.2.1.
- The CD-i disc was used in all units subject to the test.
- An analog VU meter (with 0.1 dB resolution) was connected to the RIGHT
analog audio output of each unit under test.
- The ABSOLUTE level of the signals was not measured and varied somewhat
from unit to unit.
- Only the RELATIVE amplitude between the three assets were measured,
with the ADPCM level on each respective unit used as 0 dB reference.
MediaMogul 2.2.1 Disc:
Unit DV Cart MPEG Level (dB) MPEG Level (dB) ADPCM Level (0 dB)
(MPEG #1) (MPEG #2)
210/25 22ER9141/37 0 +4 0
220/37 22ER9141/00 0 +4 0
220/57 22ER9141/00 0 +4.5 0
350 portable docking unit 0 +4 0
370 portable DV01 -0.5 +3.5 0
550 DV01 0 +4 0
605 (carrier) new +0.5 +4 0
605T new +0.5 +4.5 0
PC-I2M daughter +7.0 +11 0 (using OptImage CDIPlayback 2.2.1)
MAC-I2M daughter +7.0 +11 0 (using CDIPlayback 1.5, ADPCM att=6%)
MAC-I2M daughter -0.5 +3.5 0 (using CDIPlayback 1.5, ADPCM att=0%)
Conclusions:
- If MediaMogul and/or the DV plug-in were faulty with audio levels
or controls at some version, it is has since been fixed and not a factor.
- OptImage has an incomplete compliment of CD-i players. However,
given the wide disparity in age and models, the tests provide reasonable
confidence. It would appear that the ADPCM and MPEG discrepancy is ASSET
PRODUCTION related. ADPCM production tools permit easy control of amplitude,
plus the audio industry is very familiar with amplitude matching and calibration
procedures in all equipment in the production chain.
- The same CANNOT be said of the MPEG service industry. For Delta V
systems, there is little provision for monitoring or calibrating the audio
encoding equipment for a known and repeatable amplitude entering the encoding
system. Moreover, the encoding systems themselves provide NOTHING for checking
amplitudes on the encoded side so that calibration can be achieved
- The 4 dB relative difference from MPEG #1 to MPEG #2 was expected.
MPEG #1 was recorded at -6 dB absolute, while MPEG #2 was produced at
approximately -1.5 dB absolute (as best we could read the LED). This yields
an approximate 4.5 dB difference between the two MPEG assets--confirmed
in the tests
- The level discrepancy problem ACROSS PLAYERS, reported last October
with MediaMogul 2.2, appears to have been solved by MediaMogul, DV Plugin
and Script2Disc versions 2.2.1. The relative levels between the three
assets remained essentially identical across units
- The history of the "6%" attenuation setting in CDIPlayback/I2M
software is unknown, but apparently misguided. The 0% setting yields much
better amplitude matching with controlled assets. Unfortunately, the PC
version of OptImage CDIPlayback does not permit adjustment of the ADPCM
attenuation--it is fixed at 6%
- Similar tests using the three assets, but under Balboa, yielded essentially
identical results of the MediaMogul disc. However, on PC and Mac I2M cards,
it seems that Balboa is manipulating the ADPCM attenuation. On the PC,
the ADPCM level was only 1 dB below MPEG #1 (compared to 7 dB below with
the Mogul disc). On the Mac, the ADPCM level was 7.5 dB below MPEG #1
(same as Mogul disc), but REGARDLESS of CDIPlayback attenuator settings.
Recommendations:
- Locate equipment or software that permits monitoring of encoded MPEG
audio levels in Real Time or in Post processing
- Locate equipment or software that permits calibration
of the A/D converters and audio encoding equipment against an in-line analog
or electronic VU meter. This would permit manual level adjustment of the
Video Tape machine or other sources, to produce known and repeatable levels.
The Delta Vx system just so happens to use an outboard A/D converter with
an LED to indicate -1.5 dB absolute level. For Delta Vc and other totally
integrated systems, there is no published relationship between volume settings,
analog signal levels vs. digital levels, etc. Empirical testing--effective
but tedious-- could be used to calibrate the system
- Alternatively, process all audio files using a software audio MPEG
encoder. Amplitudes can be controlled/manipulated in standard audio production
tools, with seemingly predictable encoding results
- For a given title, both the ADPCM processing chain and the MPEG processing
chain MUST be calibrated to the same reference amplitude. Without calibration,
the audio levels between ADPCM and MPEG assets will be unpredictable.
Most audio experts recommend a "0 dB" reference at an absolute
level of -10 or -15dB (absolute 0 dB is full 16 bit modulation) to provide
adequate safeguards against clipping. If MPEG and ADPCM processing cannot
be mutually calibrated, then at minimum, the calibration references should
be published by both so the title developer can make the appropriate adjustments.
Home Page | Up
One Level | Site Map | Send
Us email