Testing, testing…
Easy steps in improving your studio sonic performance
By Dino
Ziogas (additional editing by Bob Dennis)
One of the main focuses of my previous
article (Money for Nothing – a general guide on equipment
upgrading, REQ May issue, 2004) was this simple principle:
What the home recordist can’t really afford is to underuse their
equipment! In that article one of the things I did was to offer
some guidelines on optimizing the range of uses that can be had by
using the already available tools. This installment will focus on
another, equally important, aspect of equipment performance: signal
integrity. The studio environment offers a wide margin of
improvement in noise figures, hum presence, distortion levels and
line-up settings and it is exactly these potential trouble areas
that we’ll try to tackle using simple, easy-to-find tools. Most of
these incorporate the use of the omnipresent standard home-PC and
guess what, even not powerful computers can handle the testing tasks
really well.
So, get your lab coats and off we go!
Test Tones
Many testing, measurement, calibration and
line-up procedures require the use of a reference or test tone fed
through the tested equipment. Before getting into what these
procedures are all about let’s see some facts about the test tones
themselves in order to acquire a better understanding of why things
are the way the are.
The most common
test tone around is a 1khz sine wave. It consists of a
fundamental @1khz and no harmonics (pure sine wave). Thus
if anything in the audio
chain is non-linear or distorts the signal,
the distortion is usually
very obvious almost
immediately.
The 1khz frequency is used just because it’s an easy number and in a
logarithmic frequency scale sits in the middle of the audible
spectrum! There are uses for sine waves of other frequencies but
we’ll discuss them later.
Creating such a
test tone is particularly easy.
For a tone generator I'm
using my digital Korg X5D synth
in single
oscillator mode set at sine
wave and normal octave (8’),
making sure the
oscillator
level is nearly full, there are no pitch, volume or filter
modulations and of course no built in FX (the sinewaves may not be
particularly pure in the average synth but
for testing purposes they
are perfectly good). Now the
1khz tone is the C key two octaves above middle C (it’s actually
1037hz but this in practice imposes no problems!). Also, most
audio editing
software provide such generator
facilities. In Cool
Edit Pro (now Adobe
Audition) there’s
the Tones option in the Generate menu,
in
Sound Forge
it’s
the Simple Synthesis option in the Tools menu, and in Wavelab
use
the Audio Signal Generator in the Analysis menu
(to name but a few).
Set them
for a sine wave at
0dBFS (or any other intended
level), a frequency
of 1kHz and 16bit/44.1khz
stereo. The duration of the tone can be anywhere from a few seconds
and up to a minute depending on the use. To avoid large files you
can take the 5sec file and set the audio player for loop playback.
Alternatively, there are many ready-to-use files on the web. For
example go to
www.rme-audio.com
and to
the download section.
Another useful test signal is white noise (a random
energy where there is an energy distribution so that the amount of
energy is the same for each cycle, causing the noise level to
increase with frequency – taken from RID/REQ glossary). Various
software packages provide it as well as synth patches but I wouldn’t
trust digital solutions that much because digital ‘white noise’ is
in reality a pseudo version of the real thing (digital randomness is
not that random!). Especially for digital synths, avoid to use noise
patches that are ‘pitched’. An analog white noise generator is a
‘better’ solution but don’t let this fact deter you from trying out
the digital variety.
More complex tones exist as well (usually
sequences of simpler tones) but I won’t go into these as in most
cases they are automatically generated and performed by the testing
software and this article would become more tech-orientated than
intended!
Anyway, the
good news is that there are ready-to-use test tones CDs in the
market so there’s no need to fiddle around with generators or
software to get on the testing route if you don’t feel like it.
Get in line
Signal to noise
ratio is often compromised by improper alignment between equipment
and a trouble spot is the signal exchange between the desk [mixer]
and the recording device. Due to the current technology status
today’s recordists use mostly digital recording machines and as a
matter of fact this very combination of analog (e.g. desk) and
digital devices is a frequent usual cause of misalignment.
It's
all related to the inherent problems of interfacing analog and
digital equipment (signals) when an A-D or D-A converter is present.
Balanced analog signals have their nominal level at +4dbu (a
convention). Analog systems typically cater for the occasionally
peak up to +22dbu (some high-end systems
can go up to +26dBu), therefore
exhibiting a 18db (or 22db) of headroom. Digital systems are
configured in another fashion. Their zero point (0 dbFS - FS meaning
Full Scale) is the upper limit for digital signal amplitude. There's no headroom above that. Digital
systems cannot tolerate overload so a safety margin had to be
adopted.
In digital
conveters (A/D or D/A), the most used nominal level
is a
1000 Hz test
signal to translate
to -18dbFS in the digital realm. This level is used because the usual
headroom for the analog signal and the headroom for the digital
system are matched at 18 dB. Headroom is the difference, in dB
between reference level and the maximum possible level you can have.
This "headroom allows higher level transient peaks that wouldn't
register on a VU meter to get through without distortion. To find more on the decibel
check out the glossary section at the end.
Aligning mixer to recorder.
First decide on the required headroom and lets say you’ve decided to
use 18db. Then send a test tone and raise the master faders until it
reads 0 (+4dbu) on the console meters (I assume you have already
aligned the soundcard to the mixer if the soundcard is to be used as
the test tone source) and set the recorder in input monitoring mode.
Now raise the input trim control of the recorder until the recorder
meters read –18dbFS.
Notes:
One
would expect that since the meters on the console (and on the
majority of today’s designs) are peak-reading you could rely on
these for. In truth, peak meters (of the hardware variety) can still
under-read the briefer transients by as much as 6db as a result of
their design. In practice you should use the converter’s or
recorder’s meters to be sure and if you expect transient-heavy
material through the desk simply keep the peaks at a lower LED on
the mixer meters as a quick reference.
Something further to keep in mind: when the
player’s confidence builds up his/her output will be usually about
2-3db higher, which is something that must be considered during
soundcheck.
More
alignments!:
- Soundcard
to mixer.
Feeding a
test tone at 0dBFS produces
the maximum output from the converter, so you can adjust the
soundcard return
mixer channel input gain
knowing that there’s a level
ceiling that will not be exceeded. Just make sure the mixer meters
barely flicker at the first peak LED (in some mixers this may result
in slight distortion so use your ears).
- Matching
of mono channels for stereo work:
Some budget mixers have gain variations between their channels as a
result of less than ideal quality control and materials used. When
playing back stereo signals through two mono channels
use the test tone to
match
the channels’
input gains. This way the
stereo field will remain unaffected. This is also a good way to
check part of your mixer’s quality.
- Analog
tape calibration. This
topic is too extensive to be included here but I advise interested
readers to check out the following link on the subject:
http://www.recordingwebsite.com/articles/tapecal.php
Careful when crossing over (or the unspoken benefits of white
noise testing)
Crossovers can be really useful tools but in
more critical tasks (such as mastering) they have to exhibit minimal
phase shift. Here’s a way to test your crossover unit to check if
it’s really up to the task.
1.
Set EVERYTHING at unity gain & vary the crossover points till
they almost overlap in the center (let’s say around 1K) using
white noise and again later with complex audio like a hard rock
band.
2.
Listen for brittleness. There should not be a change in
sound quality until the crossover points are within maybe 1/3
octave.
3.
Compare that sound to the source in bypass mode (You can set
the mixer up so you can switch back & forth between the processed &
unprocessed signal with a single pushing of a button – use the
various routing features of your mixer or set up a special patch).
Note: Crossover testing tip by Silent
Bob!
Analyze this!
Ears may be a
great tool but they do have limitations and only when used together
with a visual representation of the frequency content they can
really be used for certain troubleshooting tasks. The most common
tool for this kind of job is a spectrum analyzer. I have
found some excellent freeware and I’d definitely recommend the
Wavetool bundle by Paul Kellet or the VisualAnalyser (VA)
5.5.3 by Alfredo Accattatis (for something with more features) not
to forget the acclaimed RightMark Audio Analyzer (RMAA) – now
at version 5.3. The above utilities don’t limit themselves to a
spectrum analyzer but contain other useful tools as well. There are
many sites where you can download various programs of this type – I
have used www.hitsquad.com/smm.
Before seeing what you can do with a spectrum
analyzer here’s a short description of some of the controls (some
programs may have less or more parameters depending on their level
of sophistication). Please note that I chose not to get into too
much depth as it is beyond the scopes of the present article.
·
FFT size:
This affects the display resolution – to put it simple: The larger
the number, the more detail you get on the graph. Unfortunately
bigger sizes affect the real-time performance of the program and
whereas for general applications sizes of 8K-16K (8192 to 16384) are
fine, sometimes you may have to lower the value to something like
2048 to get unaffected real-time tracking of the incoming data.
However if real-time analysis is not a priority (say you need to
examine a small portion of a track or do some restoration analysis,
etc) you can keep it set at a big size.
·
Smoothing window
algorithm: This parameter
basically determines how sharp the peaks appear on the graph. Common
options are ‘no smoothing’, ‘Rectangular’, ‘Hanning’, ‘Blackman’,
Bartlet’, ‘flat top’, etc. The first two are better at helping the
identification of hum, whistles etc as they make the peaks look
sharper (but are less accurate). For general mix analysis,
‘Blackman’ (or ‘Blackman-Harris’) is more suitable.
·
linear or
logarithmic (dB) level scale:
In Linear peaks show up clearly. Logarithmic (dB)
is useful for making measurements, especially of low level or wide
range signals. The human ear perceives sound in a similar fashion so
the logarithmic scale is useful for getting a first general
impression of the signal.
·
Octave band display
or Octave grouping: The particular parameter determines the
number of bands the audio spectrum is divided in. As you probably
suspect more bands means better resolution but with a toll on
performance. One of the better compromises is 1/3 octave. Again, the
application determines the need for one setting or the other. I’d
suggest the smaller setting (i.e. 1/6 smaller than 1/3) that doesn’t
appear to affect the performance or make the display too confusing
to read.
·
Averaging:
This smoothes the display by averaging over N (= the
parameter value you entered) spectra. Better left at small numbers
(1, 2 or 3) but again try the smallest value that doesn’t compromise
the performance noticeably, if smoothing helps your needs.
…keep the hum please
Hum-infestation
is so common that some times I wish I had a penny for every amp I’ve
heard exhibiting this kind of interference! The really bad side of
the story is that the hum gremlin doesn’t limit its presence on
instrument amps but can end up on your multitrack recordings,
monitoring system etc. Turning up the volume of the monitors may or
may not reveal the problem but this is generally not recommended as
it can be harmful to both your ears and the monitoring equipment if
the speakers are left at high volumes while you change cables to
determine the cause of the problem!
Hum is usually
caused by poor grounding or general ground loop problems and can be
really audible when unbalanced connections are used due to their
inherent absence of noise rejection capabilities. The more
complicated the studio setup the more chances there are for such
interference to step in as the possibilities for ground loops are
greatly increased. My recommended testing method would be to use
your analyzer to check, from the console main outs, all the signals
at the inputs and generally any other source you suspect of hum or
wiring problems. Routing signals through the console is not
compulsory but helps to identify problem from the interaction of the
console with the rest of the equipment (usually the mixer gets used
throughout the production!). People with no hardware mixers can
monitor their soundcard or other outputs of course!
Here’s an
example of identifying a hum problem. The software used was
VisualAnalyser.
I took a stereo
feed from the aux outs of an older portable radio/cassette player
(those with the 2-pin slim power plugs) and sent it to my soundcard
line in. By clicking on the On button on the VA window
and while having the radio outputting no specific signal (so nothing
could mask any interference) I was able to get the following
reading, shown in Fig. 1 (I’ve used the Edit Spectrum option in
VA to acquire a still of the analyzer graph window).
Hum is very
easily recognized as spikes on 50Hz (remember, I live in Europe so
the fundamental is at 50, in the US it’s at 60Hz), 100Hz, 150Hz,
200Hz etc (the harmonics). I deliberately set the frequency axis on
logarithmic scale to exaggerate the lower frequencies and the
vertical axis to show –80dbFS as max for easier identification of
spikes (not that it was needed, but again this is an educational
article!).
As a bonus here’s some first-aid tips to
minimize hum and ground-loops related problems. Difficult situations
may need more drastic (read: demolishing or re-wiring!) solutions,
but these are a great first step:
·
Believe it or not some
house sockets will cause noticeable louder hum than others – it’s a
construction issue that usually has to do with less than ideal
craftsmanship and the fact that grounding methods in house-making
are usually not that effective for audio equipment uses. Use the
analyzer to find the available socket that exhibits the less hum
level.
·
Where possible use
balanced cabling and usually a good quality audio-specific
transformer or DI box helps when interfacing balanced and unbalanced
sources.
·
Draw all studio
equipment power from the same wall socket. Failing to do so is a
frequent cause of ground loops as different sockets can exhibit
different grounding voltages (ideally it should be at 0Volt for all
sockets but this doesn’t happen in practice). Don’t worry about
excessive power consumption issues – most sockets can provide up to
1500watt without problems. (exception to this
all-in-one-socket rule is the PC used for audio purposes, this
should be powered from a different socket than the restt!).
·
Try not to have one
extension multi-socket cascading from another if you can help it.
It’s another cause of ground loop problems. Also keep the sockets in
good condition and dust-free. From time to time spray some contact
cleaning fluid to the ground plates and unscrew them to service any
loose contacts inside (with the power off, of course!).
Balancing acts
One of the
deficiencies of mixing on nearfield monitors is that you don’t hear
properly the lower octave and thus you can’t really tell how the
deep bass is on your recordings. This poses some great risks in how
the material will sound when transferred to a system which can
reproduce those low frequencies. Moreover, too much low energy may
damage the cones on smaller speakers. Adding a sub to your setup
makes a great improvement but what if the addition of the sub is
non-practical/too costly/unfamiliar etc? Don’t worry, your old
friend the spectrum analyzer can be of great assistance! With it you
can see how much deep bass there is and compare it to that of
similar-style commercial recordings. From now on you can make any
adjustments you consider will enhance the mix.
Comparing your
mix to a similar commercial can also help see what frequency balance
is used on the CD and so it’s easier to achieve the desired sound.
The beauty of it all is that any tweaks and changes are immediately
visible on the screen (provided you feed the analyzer from an
appropriate output – as I said earlier the mixer main outs are
usually more helpful).
Do you remember
when you tweaked your compressor settings and weren’t sure how
exactly it affected the sound or what frequencies/transients are
treated? With the analyzer the answers are right in front of your
eyes. Untamed transients and uneven frequency compression stick out
like a sore thumb!
Sonorous graphs
When I got into
analyzers I said to myself: “That’s all I ever needed to see
everything about a signal! How have I lived without it for so
long!”. I was, of course, wrong… There is a tool that is more
helpful than the analyzer in identifying whistles, periodical
changes of frequency content or other come-and-go anomalies and the
low-level hum and DC-offset gremlins. Welcome to the wonderful world
of the sonogram.
A sonogram
basically displays frequency over time but the twist here is that
the relative levels of the various frequencies are shown in
different colors. Sonograms exist either in well-known commercial
software like Cool Edit or Steinberg Mastering Edition
but there are some great freeware ones like that in the Wavetool
bundle discussed earlier.
Here’s a
run-through of the more basic controls (parameter names may vary in
different software but you’ll be able to identify them):
·
Resolution:
The number of bands to split the spectrum into. Each band is drawn
on one row of pixels. The setting depends on screen size and desired
detail level. A setting of 512 is usually fine. A finer detail graph
takes more processing power and time as you can imagine.
·
dB re. Full Scale:
Audio at this volume level will be drawn with the 'loudest' color.
This parameter is pretty straightforward I think!
·
Range:
The range of levels to be represented by the full range of colors.
Some time using the linear mode or a small db range (no more
than 40) may not reveal enough problems – particularly in older
analog or noisy recordings. Some times a figure of 80db may be
needed to fully appreciate the condition at hand.
·
Color Pattern:
The most usual is the greyscale representation of the different
levels but other options exist, including blue to red or the Cool
Edit-like ‘heat’ pattern.
·
Smoothing or
Hanning window: Each
frame (time slice) of audio is 'faded out' at the beginning and end.
Without this, unwanted streaks may appear in the graph.
To help you get a general idea of this tool lets see the next
example. A while ago I was sent a piece of an audio file containing
an old noise recording of a baseball game that exhibited some
come-and-go noise swells that made the sportscaster’s voice hard to
hear at times. To be able to offer some first advice on the sender
(to see if professional restoration was needed, something I couldn’t
perform) I run the .wav file through my analyzers and
a sonogram. Shown in Fig. 2 is the result I got from the sonogram. I
remind you that the horizontal axis represents time and the vertical
frequency (low freqs at the bottom). To give an indication of the
frequency axis scaling (which unfortunately is not visible on this
window as the program I used lets you see the frequency and its
level by clicking on the screen), I’ve inserted some rough numbers
on the graph.
The noise
swells are evident as a semi-periodical blurring on the lowest part
of the screen (800-4000hz) covering much of the vocal range. Also
there isn’t much going on above 8000hz which is totally reasonable
since sound came from an early ‘80s VCR. When interpreting a
sonogram the continuous horizontal lines are usually the ‘offenders’
like hum, whistles, DC-offsets etc. In our example the horizontal
line at ~15700Hz is probably a DC offset (which can easily be
notched out). The line is identified as a DC-offset because it is of
too high frequency to be hum or whistle. To understand what a
DC-offset is go to the glossary at the end of this article.
Word of
caution: Interpreting
the analyzer and sonogram requires some experience and know-how.
Practice and read as much as you can about the subject and you’ll
soon be rewarded.
Oh no! There’s more?
My original
intention was to further expand this article to include some
advanced uses of the aforementioned tools but this has grown so big
that I’ve decided to break it down into two parts. The next
installment will probably be on the next REQ issue but I don’t
promise anything!
Because I’m a
man who usually (he, he) practices what he preaches, this
summer I’ve set up a cheap customized PC (I’ve even put it on
wheels!) specifically for testing and trouble-shooting studio
issues. Since I’m in the middle of reforming my recording space this
came in very handy. An update on that will be given with the next
article as temporary technical and financial problems prohibited the
smooth operation of the system…
Cheers,
Dino
Figures
Figure 1. Analyzer
example – hum problem
Figure 2. Sonogram
example – a DC offset and periodical noise swells
Glossary
Hum
- The 60 (or 50) Hz power line current
accidentally induced or fed into electronic equipment.
hum components
The harmonics of the AC mains supply. The Americas (except the
southern half of South America), Japan, Taiwan, Korea and the
Philippines use a 60-Hz system, placing the most annoying 2nd and
3rd harmonics at 120 Hz and 180 Hz. For Europe, and the rest of the
world using 50-Hz mains, these components fall at 100 Hz and 150 Hz.
white noise
1. Physics. Analogous to white light
containing equal amounts of all visible frequencies, white
noise contains equal amounts of all audible frequencies
(technically the bandwidth of noise is infinite, but for audio
purposes it is limited to just the audio frequencies). From an
energy standpoint white noise has constant power per hertz
(also referred to as unit bandwidth), i.e., at every
frequency there is the same amount of power (while
pink noise, for instance, has constant power per octave band
of frequency).
decibel Abbr.
dB Equal to one-tenth of a bel. 1. A
measuring system first used in telephony where signal loss is a
logarithmic function of the cable length. 2. The preferred
method and term for representing the ratio of different audio
levels. It is a mathematical shorthand that uses logarithms
(a shortcut using the powers of 10 to represent the actual number)
to reduce the size of the number. For example, instead of saying the
dynamic range is 32,000 to 1, we say it is 90 dB [the answer in
dB equals 20 log x/y, where x and y are the different signal levels].
Being a ratio, decibels have no units. Everything is
relative. Since it is relative, then it must be relative to some
0 dB reference point. To distinguish between reference points a
suffix letter is added as follows:
0
dBu Preferred
informal abbreviation for the official dB (0.775 V); a voltage
reference point equal to 0.775 Vrms.
[This reference originally was labeled dBv (lower-case) but
was too often confused with dBV (upper-case), so it was changed to
dBu (for unterminated).]
+4
dBu Standard pro
audio voltage reference level equal to 1.23 Vrms.
0
dBV Preferred
informal abbreviation for the official dB (1.0 V); a voltage
reference point equal to 1.0 Vrms.
-10 dBV Standard
voltage reference level for consumer and some pro audio use (e.g.
TASCAM), equal to 0.316 Vrms. (Tip:
RCA connectors are a good indicator of units operating at -10
dBV levels.)
0
dBm Preferred
informal abbreviation of the official dB (mW); a power
reference point equal to 1 milliwatt. To convert into an equivalent
voltage level, the impedance must be specified. For example,
0 dBm into 600 ohms gives an equivalent voltage level of 0.775 V, or
0 dBu (see above); however, 0 dBm into 50 ohms, for instance, yields
an equivalent voltage of 0.224 V -- something quite different. Since
modern audio engineering is concerned with voltage levels, as
opposed to power levels of yore, the convention of using a reference
level of 0 dBm is obsolete. The reference levels of +4 dBu, or -10
dBV are the preferred units.
0
dBr An arbitrary
reference level (r = re; or reference) that must be
specified. For example, a signal-to-noise graph may be calibrated in
dBr, where 0 dBr is specified to be equal to 1.23 Vrms (+4 dBu);
commonly stated as "dB re +4," that is, "0 dBr is defined to be
equal to +4 dBu. -
This is common in many
peak reading meters on modern
consoles.
0 dBFS
A digital audio reference level equal to "Full Scale." Used in
specifying A/D and D/A audio data converters. Full scale refers to
the maximum peak voltage level possible before "digital
clipping," or digital overload of the data converter. The Full Scale
value is fixed by the internal data converter design, and varies
from model to model.
Crossover (Crossover
Network) - A set of
filters that "split" the audio signal into two or more bands
(two or more signals, each of which have only some of the
frequencies present).
Crossover Frequency
- 1) The frequency that is the outer limit of one of the bands of a
crossover.
DC-offset A DC offset occurs
when the oscillating signal ground reference (‘center’) line is no
longer a ground potential (zero volts), but has shifted slightly in
one direction or the other - anything from a few millivolts to a
significant portion of a volt. The audio signal isn’t affected by
this offset until it (the signal) is of sufficient amplitude to
approach the maximum levels, where one half of the wave form will
hit the rails before the other half, because of the offset and
premature clipping on half of the audio waveform may occur. In the
digital world, a DC offset is usually caused by a poorly aligned (or
badly designed) A-D converter. It used to be very common indeed back
in the late 1980s, but is relatively rare these days. It can also be
caused by misalignment or failed components in the analogue world
too.
(NOTE: all terms
except “DC-offset” edited from RID/REQ glossary and
RANE Pro Audio Reference. I would like to express my thanks
to Mr. Hugh Robjohns for his help by explaining the DC offset
concept, to ‘Silent Bob’ for his crossover testing tip and above all
to Bob Dennis for his insightful comments!)
|
