Active Noise Cancellation With TMS320C5402 DSP Starter Kit

locketparachuteElectronics - Devices

Nov 15, 2013 (3 years and 11 months ago)

68 views

Active Noise Cancellation With TMS320C5402
DSP Starter Kit

Farrukh Nagi, Jennifer Low S.K, Seet Li

University Tenaga Nasional

Km 7, Jln Kajang
-
Puchong

43009 Kajang, Selangor

Email: farrukh@uniten.edu.my


Abstract


The paper describes the use TMS320C5402
DSK for
single channel Active Noise Cancellation (ANC) in duct
system. The canceller uses a feedback control topology
and is designed to cancel narrowband periodic tones.
The signal is processed with well
-
known filtered
-
X Least
Mean Square (filtered
-
X LMS)

Algorithm in the Digital
Signal Processing. The paper describes the hardware
and use chip support libraries for data streaming. The
FXLMS algorithm is written in assembly language
callable from C main program. The results obtained are
compatible to the ex
pected result in the literature
available. The paper highlights the features of
cancellation and analyzes its performance at different
gain and frequency.


Keywords: Active noise cancellation, acoustics noise,
system modeling, identification


1 INTRO
DUCTION


Ambient noise has become a significant
problem in environments where industrial equipment
such as engines, air
-
conditioning systems and
transformers are used. Exposure to high decibels of
sound proves damaging to humans from both a physical
and a
psychological aspect [1].


Active noise works on the principles of destructive
interference between the noise wave generated by
primary source (the on to be cancelled) and the
interference wave, generated by the control source
(usually a loudspeaker). It
is the control and generation
of this acoustic inference wave that is processed by the
Digital Signal processing (DSP). Active Noise system
allow for attenuation of narrowband noise and low
frequency noise.


Two well
-
known control of ANC are feedback and
f
eedforward. In the feedforward topology the reference
noise and cancelled noise are used in contrast to
feedback in which only cancelled noise are used at
primary noise to estimate by the canceller.

Earlier ANC





techniques utilized fixed point DSP TMS3
20C5x[2] with
feedforward control requires extensive assembly
language programming proficiently. An expensive but
efficient TMS320c54x EVM [3] with feedforward
control can gives better results. The work describe in this
paper uses less expensive TMS320C540
2DSK with
feedback ANC control using mixed C and assembly
efficient coding and libraries to obtain comparative
results.


2 FEEDBACK ACTIVE NOISE CONTROL


The basic idea of an adaptive feedback ANC is
to estimate primary noise and use it as a referen
ce signal
x(n) for the ANC`s LMS
-
filter. In Figure 1, the primary
noise is expressed in z domain as D(z) = E(z) +

S(z)Y(z)


where E(z) is the signal obtained from the
error sensor and Y (z) is secondary signal generated
from the adaptive filter . S(z) is
the transfer function of
secondary
-
cancelled path assumed to be same as P(z) is
estimated by S^(z)

S(z)


P(z). The estimate d^(n) of
the primary noise d(n) is a synthesized reference signal
x(n)that is


X(z) = D^(z) = E(z) + S^(z)Y(z)


The secondary si
gnal y(n) is filtered by the secondary
-
path estimate S^(z) and then combined with E(z) to
regenerate the primary noise [2]. The single channel
adaptive feedback ANC system using the FXLMS
algorithm is illustrated in Figure 2, where S^(z) is also
required t
o compensate for secondary path. The
reference signal x(n) is synthesized as



x(n) = d^(n) = e(n) +



where s^
m
, m = 0,1,…..,M
-
1 are the coefficients of the
th
M

order FIR filter S^(z) used to approximate the
secondary path transf
er function [2].









1
0
)
(
M
m
m
m
n
y
s








DSP Control Domain

Cancellation zone

S(z)











Acoustic Domain















Fig.1 Adaptive Filter System Using FXLMS


Algororithm



3 EXPERIMENTAL SETUP


A large portion of ANC has been directed toward the
active control of acoustic nois
e in ducts such as heating
ventilation, and air conditioning (HVAC) or exhaust one
ducts because of widespread potential industrial
applications and the feasibility of building experimental
systems. A test rig is constructed using an 8
-
inch PVC
duct with Y

branch to accommodate two loudspeakers
[2]. One loudspeaker is used to generate simulated noise
while the canceling noise channeled through the other
speaker. A microphone is placed in cancellation zone to
pick up the resultant noise signal, which is feed
back to
ANC algorithm for computing the filter coefficients.
DSK TMS320C5402 calculates signal from the
microphone and generate an anti
-
noise signal for an
amplifier at the next stage from their D/A. The
experimental setup of the duct is shown in Figure 2.

The



dimensional set up of the duct system is decided based
on factors including sound frequencies of interest and
physical spaces available.




















Fig. 2 Duct System



4 SPEAKER
-
MICROPHONE SETUP


For the can
cellation to work well, the distance of the
microphone and the loudspeaker should not be changed
once the identification of the secondary transfer function
path has been set. The system transfer function is
increasely difficult to equalize as the microphon
e moves
away from the speaker (the phase change gradual over
the frequency of interest to very rapid). An amplifier
should be able to drive the speaker to generate the same
sound pressure at the microphone as the noise itself
without control.



5 LEAS
T MEAN SQUARE (LMS) ALGORITHM


A least mean square (LMS) approach is widely used for
adaptive filter routines. This algorithm does not require
squaring, averaging, or differentiating. The LMS
algorithm provides an alternative method for
determining the opt
imum filter coefficients. The block
diagram of an adaptive FIR filer is shown in Figure 3.

S^(z), estimate
of S(z)

y(n)

d^(n)

W(z)

LMS

S^(z
)

x(n)

y(n)

+

Unknown
Pr
imary
Transfer
Function,
P(z)

Secondary
Transfer
Function,
S(z)

d(n)

y^(n)

e(n)

+

-

Speaker
mounting


Noise
speaker

DSP

TMS320C5402

0.64m

0.52m

NOISE
SPECTRA
LAB
















Fig 3 Adaptive Filter Using Least Mean Square



The least mean square (LMS) algorithm, which is given
by







where




e(i) = d(i)


y(i)









The output of the adaptive filter is given by










6 DUCT SYSTEM SECONDARY PATH


MODELING



In real life situation, secondary path has enormous
impact on the ANC system design. In fact, secondary
path will effect the response of the AN
C filter and the
convergence of the ANC algorithm. The FXLMS
provides an understanding of the impact of the
secondary path effect S(z) on the proper functioning of
an ANC system. The design of feedback ANC system
should incorporate this effect. Offline mod
eling
technique can be used to estimate the secondary transfer
function S(z) during the training stage assuming the
characteristics of S(z) is unknown and is time varying
due to the effects of aging of the loudspeaker, changes in
temperature, and air flow
in secondary path [5] . The
estimate of the secondary transfer function S^(z) at the
end of the training period is used for the active noise
cancellation system. The system used to estimate the
secondary path transfer function is shown in Figure 4.








































)
(
)
(
2
)
(
)
1
(
k
i
x
i
e
i
b
i
b
k
k










1
0
)
(
)
(
N
k
k
k
i
x
b
i
y
y
(n)

Anti Aliasing Filter


S^(z)

d
(n)

E


(n)

Pre
-
Amplifier

Power

Amplifier

Low Pass
Filter

D/A

A/D

LMS

White noise
generator



x
(n)

Fig 4 System Identification Using Off Line Modeling


Techniques

The primary noise is estimated as





where (i = 0,1,..M
-
1) is the
coefficient of
the secondary
-
path estimation filter S(z) and M is the
order
of the filter S(z). The off
-
line modeling
procedure is summarized as
follows:


1.

Generate sampled white noise signal x(n)

2.

Obtain desired signal d(n) from error sensor.

3.

Apply adaptive filter algorithm as to compute
adaptive filter output and error signal and
update coefficients using the traditional LMS
algorithm.

4.

Run until ad
aptive filter S^(z) converges to
optimum solution.


After the offline modeling is completed the secondary
path transfer function S(z) is estimated. The estimated
coefficients for the transfer function secondary path S(z)
is shown in Figure 5 using step siz
e,d_mu values.














Fig 5 Estimate Secondary Path S^(z) coeffiecients with


step size 10


7 ADAPTIVE NOISE CANCELLATION


Physical implementation of the adaptive feedback ANC
requires both hardware and software that are capable o
f
real time signal processing. For this purpose, a dedicated
digital signal processor is the most technically practical
and economically viable method.


TMS320C5402 DSK has mono input AD50 codec, and
is good enough for single channel, feedback ANC
contr
ol. The board is equipped will two multichannel
-
buffered serial port McBSP for direct (DMA) data
transfer to and from the CPU. Chip support library, CSL
are available in ‘C’ to handle the interrupt, initialization,
and of calling ISR in ‘C’ programming. Th
e actual data
processing is carried out in assembly program residing in
interrupt service routine, ISR, which make use of the
circular buffer and LMS instruction. The data input e(n)
and the output y(n) has buffer of 64. Having small data
input buffer make
s DSP processing less demanding by
increasing throughput of the DSK. Circular buffer are
use to synthesized x^(n) and y^(n). LMS instructor
provide compact and efficient output y(n) and update the
LMS filter weights(z).


7.1 ANALYSIS AND RESULTS OF



CANCELLATION


DSK AD50 codec have programmable gain and
sampling frequency. Since LMS algorithm and two
convolutions process are heavily taxing on the DSP,
lowest sampling rate of 2kHz and 0db gain are used to
avoid the overflow and saturation. I
n LMS algorithm the
step size, (mu) control the convergence of the algorithm.
The value of 10 is best compromise for the speed and
accuracy of the convergence.



The feedback ANC is high subjective to the gain of
canceling speaker amplifier figure 2. The A
D50 output
gain was minimize in addition to left shift of output to
AD50 in assembly program. The location of the error
microphone is critical as it can easily feedback the
cancellation noise and loosing control of the cancellation
process. The result of s
ingle and multitone frequency
cancellation are shown in Figure 6
-
8.






















1
0
)
(
)
(
)
(
M
i
i
i
n
y
s
n
e
n
x
i
s
---

Without cancellation



W楴iCan捥汬慴楯n


at 130Hz

120

160

140

150

130

170

Frequency (Hz)

-
40

-
60

-
80

-
100

-
20

0

Relative Amplitude (db)

Fig 6: Cancellation result on pure tone noise at


130Hz

















































Signals with multiple frequencies are merely in low
frequency and narrow band noise signal. The frequencies
t
hat can be reduced are in range of 70 Hz to 130Hz.
Therefore the relationship between noise frequency and
noise reduction is not linear; since ANC`s ability to work
in certain range of frequency depend upon the real
acoustic parameter and PVC pipe diameter
. A good
performance of ANC system is based on its step size, mu
value. For lower step size it will give a good result but
take more time to reach convergent condition.


8.0 CONCLUSIONS


The analysis and results of experimental setup are as
expected from
the feedback ANC techniques. Limited
cancellation 20
-
25% of single and multitone noise. The
problem of regeneration of cancellation feedback are
very critical. TMS320C5402 DSK does not offer two
inputs for feedforward ANC scheme which is maximum
efficient
and x^(n) needs not to be synthesized and can
be made available to LMS via second input. For this
purpose AIC10 EVM daughter card can be used in
conjunction with C5402DSK for further work.


REFERENCES


1)

Kuo, SM.Adaptive Active Noise Control
Systems
-

Algori
thm and DSP
Implementation.Proc.:Digital Signal Processing
Technology Conference 17
th



18
th

Aug.1995,Florida.

2)

Stephane Boucher, Martin Bourchard, Andre
L`esperance, Bruno Paillard, Texas
Instruments. Implementation a Single Channel
Active Adaptive Noise C
anceller with the
TMS320C50 DSP Solutions November 1997

3)

Sen. M.Kuo, Phd, C.Chen. Implementation of
Adaptive Filters with the TMS320C25 and the
TMS320C30. Texas Instruments. September
1996.SPRA 336

4)

Sen. M.Kuo.Active Noise Control systems with
TMS320 Family,

Department of Electrical
Engineering Northern Illinois University
Dekalb IL 60115.

5)

Dr Philip C.Loizou Chair,Dr Louis R. Hunt.
Sub band Feedback Active Noise Cancellation,
The University of Texas At Dallas, August
2002


Fig 7: Cancellation result on pure


tone noise at 120+130Hz

---

Without cancellation



W楴iCan捥汬慴楯n

120

160

140

150

130

170

Frequency (Hz)

-
40

-
60

-
80

-
100

-
20

0

Relative Amplitude (db)

---

Without cancellation



W楴iCan捥汬慴楯n

120

160

140

150

130

170

Frequency (Hz)

-
40

-
60

-
80

-
100

-
20

0

Relative Amplitude (db)

Fig 8: Cancellation result on pure tone noise at


110+120+130Hz

---

Without cancellation



W楴iCanc
敬污瑩en

R敬慴楶攠䅭p汩瑵d攠(db)

Fig 9: Cancellation

result on pure tone noise at


100+110+120+130Hz

120

160

140

150

130

170

Frequency (Hz)

-
40

-
60

-
80

-
100

-
20

0