Sampling with the AWE - asking for trouble
For those who think that 16 bits always are 16 bits
Written by Mathias C. Hjelt
V1.1 13 Jun 95
Introduction
To make a long story short enough for being written in less than an hour, let me put it like this:
16 bits at 44.1kHz does not automatically mean you've got CD-quality, no matter what the PC sound card
manufacturers and advertisers say.
This stuff deals with things like why 16 bits are not always 16, and why you don't have to worry
about Nyquist's theorem when sampling with the AWE. Note: this is only about the ADC, and has
got nothing to do with the EMU and the rest of the AWE.
Frequency response
Nyquist's theorem says that the sampling frequency must be at least twice the highest frequency
of the signal that is to be digitized / reproduced. Since the AWE's maximum sampling frequency
is 45.45kHz, the highest frequency it can sample/play should be close to 22.72kHz. Unfortunately,
there are things like cheap AD converters and way too efficient low-pass filters which make
sure these frequencies never get far.
Unfortunately I don't have any fancy frequency response graphs or exact
figures to show here. Many computer mags that have tested the AWE have
reported that high end is being attenuated pretty much above 15kHz, but I
doubt that. My own brief tests showed that an 18kHz signal is somewhat
attenuated, but definitely not down by 12dB. However, if you're looking for the
ultimate high-end, this card won't be of much use.
The low end (bass, that is) is also mistreated. Somewhere between 20 and 50 Hz there's
a slope that chomps up several dBs. Below that the level is pretty constant at a few -dB. However,
these frequencies are seldom taken into account in "real" tests (in computer mags, that is) since
50Hz is usually where the graphs start.
In any case, the playback side of the AWE has got much better frequency response than this darned
recording part. But how bad is this? Does it make the card useless? There are - as always - two
answers: yes and no. For most users, no, for the audiophiles and other serious dudes, sort of.
If you just want to sample your own sounds and use them with the AWE's EMU, the recording
quality will be sufficient, and most regular music applications will work fine. But if you want to
do real accurate signal processing, research etc, you'd probably want a much cleaner signal and
a much better frequency response. But that's not what the AWE is all about, so why feel sad?
Noise fighting - an impossible but not useless task
When sampling with the AWE, make sure you use the line input, never the mic input, and see to it
that you always use recording gain x1. This does of course mean that you can't sample right off a mic,
and that you'll run into trouble when sampling stuff that has got rather weak signal level, but
instead you'll be rewarded with as good SNR (signal-to-noise ratio) as possible. Note that "as good as"
is not equal to "good".
If you want to sample something using a mike or some other low-level audio device, you need to
use an external pre-amplifier to be able to get a sufficiently high signal level for the AWE's line
input. It is general knowledge that you always should use as much as possible of a sample's amplitude
range (but yet not have the peaks clipped), and this is why you need to make sure you've got enough
volume on the input. A suitable pre-amp may be a mixer or even a guitar pre-amp. Anything that can
amplify weak signals without adding noise will do fine.
Why not use the mic input? Forget it - it's flooded with noise. Why not increase the recording gain
factor to x2 or x4? Now that's a better question. The problem is that when you increase the gain, you
also increase the background noise level. With background noise, I mean the crap that you get when you
sample stuff without any inputs enabled, or alternatively with the inputs enabled but tied to ground.
I did these tests using both methods, running at 44.1kHz, 16 bits, mono:
Gain x1:
Totally clean bits: 12 (highest noise peaks at -74.5 dB, ~ 3.6 bits)
Average noise level: About -78.5 dB (3 bits)
Gain x2:
Totally clean bits: 11 (highest noise peaks at -69 dB ~ 4.6 bits)
Average noise level: About -72 dB (4 bits)
Gain x4:
Totally clean bits: 10 (highest noise peaks at -63 dB, ~ 5.5 bits)
Average noise level: About -68 dB (5 bits)
Quite embarrasing, isn't it? The only (poor) consolation we've got is that the noise level of gain x1
won't bother that much when it is played through the EMU. It is, in fact, left out almost completely.
For an explanation of that, check out the page about why
the EMU isn't pro.
In addition to all this, the ADC has got a rather stupid DC offset error, which means that a zero
signal does not result in a zero sample. The offset isn't even the same on the left and the right channel
when doing stereo sampling. An offset error like this results in asymmetric clipping when the inputs
are overloaded, and in silent pops at the beginning & end of the sample when it is played.