Operation Manual
I-Tech HD DriveCore Series Power Amplifiers
page 62
I-Tech HD DriveCore Series Power Amplifiers
Operation Manual
page 63
1600
1400
1200
1000
800
600
400
200
0
10
20
30
40
Crossover Rolloff Rate - dB/octave
CPU Pr
ocessing Power - cycles/sample
Required CPU Processing Power
IIR
FIR
FFT FIR
50
60
70
80
90
100
Fig. 16.3 FFT convolution block diagram. here the input and filter impulse
responses are both individually FFT’d and multiplied and then inverse
FFT’d to genterate the output.
16.10 Filter Design
IIR filters are typically based on an equivalent analog circuit. Because
of this, the IIR design process is a straightforward matter of converting
an analog filter frequency response specification into digital
coefficients. In contrast, FIR frequency response is usually specified in
terms of an ideal, though physically unrealizable filter, e.g. a brick wall
lowpass. The FIR filter design is the process of generating coefficients
that approximate the response of the ideal filter. The I-Tech HD FIR
filters are designed using a proprietary iterative algorithm using
state-of-the-art signal processing techniques. It yields linear response
and predictable low-noise operation over the entire audio
frequency range.
16 Application of FIR Filters to Loudspeaker Crossovers
16.11 Low Latency 96 kHz Studio Quality Filters
Because of the extreme computational load imposed by high rolloff FIR
crossover filters, a potent DSP and highly optimized FFT Conbolution
algorithm are essential to low latency operation at 96 kHz. In-system
performance is equally dependent on the filter design process itself.
I-Tech HD’s platform-optimized FFT Convolution algorithm and
state-of-the-art filter design methodology combine to deliver low
latency 96 kHz studio quality filters.
Real World Benefits
16.12 Measurements of Two-way
Loudspeaker System
The practical advantages of using an FIR-based crossover can be
illustrated by comparing results with an IIR crossover when a two-way
loudspeaker is setup and measured. For measurement purposes, a
small two-way loudspeaker was driven by a Crown I-Tech 12000 HD
power amplifier controlled by Crown’s HiQnetTM System ArchitectTM
software. the use of a small system in this application ensures that the
frequency response and angular coverage of the individual drivers has
sufficient overlap through the crossover region to properly illustrate
frequency response and crossover effects.
The following two sets of measurements show the results of using a
conventional IIR filter set up as a 4th-order Linquitz-Riley crossover
and a FIR filter set up as a linear-phase high-rolloff crossover. Each set
of measurements first show the magnitude and phase response of the
crossover alone and then follows the frequency response measure-
ments of the system itself showing individual driver responses, on-axis
responses with the crossover in and out of polarity, and a single
off-axis frequency response where the response exhibits a null due to
driver spacing.
The small two-way system measured here has a woofer and tweeter that
are separated by about 5.5” which creates an off-axis polar null at
about ±25° at the 3 kHz crossover frequency.
Fig. 16.4 Magnitude (left) and phase (right) of a 3 kHz Linquitz-Riley
4
th
-order crossover filter.
Fig. 16.5 Measured individual driver responses (left) and overall
summed on-axis response (right) of a two-way system with a 4
th
-order
3 kHz LR crossover.
16 Application of FIR Filters to Loudspeaker Crossovers
The next two graphs in Fig. 16.5 show the measured individual driver
responses of the two-way system driven by the LR crossover as
produced by the Crown I-Tech HD amplifier and the resultant summed
1m on-axis response.
Note that the individual driver responses were first equalized using the
amplifier’s parametric equalizer to be more of less flat over a
significantly wide range above and below crossover. After the initial
equalization, the chosen crossover was applied and adjusted. Slight
delay was added to the tweeter channel to compensate for driver offset
to insure that the high- and low- pass sections were in phase
through crossover.
One unexpected benefit of the FIR crossover was the speed up of the
adjustments to insure that the drivers were in-phase through
crossover. This was because the linear/zero phase characteristic of the
FIR filter did not change the phase of the individual drivers when the
crossover was applied! The linear phase characteristic of the FIR-filter
based crossover significantly reduces the setup time of the crossover
as compared to an IIR implementation.
All measurements were accomplished using windowed freefield
techniques in a non-anechoic room at 1m using a 25 ms window. This
window size allows fairly-accurate measurements to be made down to
abut 100 Hz.
IIR Measurements
A 4th-order Linquitz-Riley (LR) crossover with 24 dB/octave slopes at
3 kHz was chosen to demonstrate a typical well-designed conventional
crossover. Although, in conventional terms, a relatively sharp 24 dB/
octave crossover slope is considered quite sharp. The crossover region
spans a relatively wide two octaves. This means that effectively both
woofer and tweeter are radiating simultaneously between roughly 1.5
and 6 kHz.
Level - dB
Frequency - Hz
20
-30
-20
-10
Low Pass
High Pass
0
10
100
1k
10k 20k
Level - dB
Frequency - Hz
20
-30
-20
-10
Woofer
Tweeter
0
10
100
1k
10k 20k
Level - dB
Frequency - Hz
20
-30
-20
-10
0
10
100
1k
10k 20k
Phase - Degs
Frequency - Hz
20
-180
-90
0
Low Pass
High Pass
90
180
100
1k
10k 20k
The following two graphs in Fig. 16.4 show the theoretical frequency
response magnitude (left) and phase (right) of a 3 kHz LR 4
th
-order
crossover. In each graph, individual curves are shown for the low pass
(blue) and high pass (red) portions of the crossover. Note that both
low-pass and high-pass filters are 6 dB down and in-phase at the 3 kHz
crossover and thus sum to unity. Note also that the phase of the low-
and high- pass sections is identical but exhibit a 360° phase rotation
through crossover. The jump in phase at 3 kHz is a byproduct of the
wrapped phase display, which keeps the phase on a ± 180° scale. The
LR responses will sum to unity at all frequencies as expected, but will
exhibit an all-pass non-linear phase characteristic.
