739
1.000
1220 0.283
1641 0.327
2464 0.346
5464 0.633
10948 0.268
191
0.193
448
0.253
1032 0.230
2186 0.201
3673 0.211
4145 0.172
6105 0.173
7110 0.153
Figure 47.3
Prominent components of turbine.
Housing and Resonances
The jet engine is uniquely interesting in this regard, as it is designed to minimise
standing waves and resonances that might cause mechanical fatigue. Although
the interior surface is smooth and reflective, the supersonic airflow means there
is no back propagation upstream. Certain speeds do accentuate turbine compo-
nents somewhat, but we can make a simplification for all but very low speeds by
ignoring the contribution of the housing. All the important sounds are emitted
to the rear through the exhaust.
526
Jet Engine
Model
Two separate models are combined. The first provides a turbine sound, which is
an inharmonic spectrum composed of separate sines obtained by analysis. The
second is a turbulent forced flame model produced from carefully banded and
overdriven noise. At low speeds the turbine sound is dominant but is quickly
overwhelmed by a very loud noisy component as speed increases.
Method
A five partial additive synthesis subpatch is used for the turbine. Analysis val-
ues are multiplied by a speed scalar, keeping the same ratio. For the turbulent
gas component a cascade of filters and nonlinear functions are used to produce
a sound that varies with speed, moving from an even low-passed hiss at low
speed to a highly distorted wideband roaring at high speed.
DSP Implementation
Figure 47.4
Forced flame.
A forced flame model adapted from a “flamethrower” patch
is shown in figure 47.4. A normalised input controls the
thrust which increases in density by a square law. The
response is obtained by taking white noise through a fixed
filter, constraining it to a middle-range band gently rolled
off around 8kHz. A very gentle filter moves over the bot-
tom range against a second fixed high-pass. This gives us a
sweep that increases dramatically in amplitude while avoid-
ing any very low frequencies that would cause “bursting” or
audible fragmentation of the sound. After multiplication by
about 100 we clip the signal harshly to create a wide band of
harmonics. This signal is passed through a second variable
filter that changes frequency linearly with the control input
between 0 and 12kHz. At low control values, hardly any sig-
nal passes the first filter stage, so the low-frequency noise is
not clipped. At medium control levels, frequencies passed by
the fixed band pass are at a maximum, so the output noise is
highly distorted and coloured. At high control levels, the clipping is less severe
again, so we hear a brighter whooshing sound.
We implement a simplification of the turbine analysis data in figure 47.5
using only five components. By experimentation, the most important five par-
tials are taken from an averaged analysis of the engine running at constant
speed over about 60s (different from the snapshot shown in figure 47.3). This
gave 3
,
097Hz, 4
,
495Hz, 5
,
588Hz, 7
,
471Hz, and 11
,
000Hz as the main frequen-
cies. To scale them, the list is unpacked and each frequency converted to a
signal so that it may be smoothly modulated by the engine speed control. A
separate
oscillator mixed in approximate ratio provides individual partials
Do'stlaringiz bilan baham: |