5
Mackie Industrial White Paper
Noise Sensing
September 2000
1.3 RMS Measurements
The nLMS algorithm requires normalized energies for calcu-
lating the FIR coefcients. A real RMS algorithm is used to
measure the signal levels [2, 4, 11-12]. This is accomplished by
buffering the squares of each sample of the input signal (Square),
averaging the sum of these squares every sample (Mean), and
taking the square root of the mean (Root). For every input sample
an RMS value is produced, which is itself buffered and ltered
by a 256-point window function, to smooth the response of the
detector by a xed known delay and amplitude.
The main benet of ltered RMS measurement is to reduce
the impact of low frequency room energy variations caused by
the music broadcast into the room, which can give variations
in the room energy readings by 2 or 3 dB. This amount of
variation could cause the compander to constantly turn the
volume of the music up and down by a similar amount, which
is quite noticeable and undesirable.
Several RMS detectors are required in the system. The nLMS
Coefcient Calculator requires the energy of the FIR output,
error and music signals. The Compander requires the level of
the music and microphone signals.
1.4 Down Sampling
To extend the effective FIR length for larger rooms the LMS
lter is down sampled from the converter sample rate [8,9]. In
this implementation, 44100 Hz (Fs) is the converter sample rate
for the music and microphone signals. Down sampling also helps
reduce the memory and MIPS requirement. This also has the
effect of comparing energies of signals ltered at half of the
down sampled Nyquist rate. As long as the noise signal has a
proportional amount of its energy in this spectrum, the ambient
noise sensor will approximate the level of the noise signal. As we
have found, this is a valid assumption.
1.5 Compander
User parameters control the operation of the compander.
Minimum Gain
,
Gain Range
,
Noise Threshold
,
Noise Range
,
Attack
/
Release Times
, and RMS measurement parameters
determine how much gain (or attenuation) is applied to the
music signal.
The compander
Attack
and
Release
parameters control the
rate at which the gain is turned up or down. These are impor-
tant to control the gain during loud sudden events such as door
slams, yells, or dropped dish trays, which would normally
cause the music level to be turned up very loud. A slow attack
rate (more than 10 seconds) with a faster release rate will
reduce the level of gain applied to the music. This allows the
compander to track the ambient room noise while ‘rejecting’
these singular events if desired.
1.6 Auto-Calibration
The biggest single problem with controlling the music gain based
on the room noise is runaway gain. Runaway gain occurs when
the compander turns up the music volume which is measured
as ‘room noise’, which tells the compander to further increase
the music volume. The compander must have the appropriate
information to prevent this cycle from occuring.
A calibrated algorithm for computing the approximate loca-
tion at which runaway gain might occur is used as an override
to limit the sensitivity to the room noise. Although the algo-
rithm constantly adapts to the room acoustics it takes a
signicant amount of time for the algorithm to get to its best
approximation of the RTF. Until this point is reached it is
not possible to determine the override conditions required to
prevent runaway gain. This condition would occur every time
the unit was reset (power on/off) were it not for the
Auto
Calibration
process, and the internal non-volatile storage of
the calibration parameters.
The
Auto Calibration
uses a real music signal with xed gain
to adapt and monitor the ambient room noise. As long as the
algorithm is making progress towards the RTF it will continue
to adapt and monitor the noise level. During calibration it
is best to have a minimum amount of room noise so that
the algorithm can determine the progress towards the RTF.
Once the progress towards the RTF has slowed signicantly
the current adaptation coefcients are stored and the Noise
Threshold Override is computed. Every time the unit is reset
the calibration coefcients and Noise Threshold Override are
restored. This allows the compander to prevent runaway gain
on reset and during normal operation. The algorithm contin-
ues to adapt at all times to keep up with room acoustic
changes, and as long as the room acoustics do not change
drastically the gain should be prevented from running away.
2 Hardware Setup
The SP-DSP1™ was designed to be an expansion card that is
added to our new “SP” or Sound Palette® series mixer/ampliers
(SP2400/1200). These ampliers provide one or two zones of
200 watts per channel in a two-rack-space package. Since the
SP-DSP1™ operates
only
on a mono program signal, one card
is required per channel. Once the card is installed, the user
needs to connect an ambient microphone to the rear panel of
SP2400/1200. Note that the connection requires a 3.08mm 3-pin
male Phoenix connector (i.e. Euroblock).
