Int. J. Electron. Commun.

(AE) 62 (2008) 438 444

www.elsevier.de/aeue

Adaptive cancellations of systematic disturbances in a synchronous

amplitude demodulator for sensors and instrumentation applications
Santanu Sarma
ISRO Satellite Centre, Control Systems Group, Vimanapura Po, Airport Road, Bangalore 560017, India
Received 15 March 2007; accepted 19 June 2007

Abstract
An adaptive technique is presented that performs synchronous amplitude demodulation while adaptively canceling systematic disturbancesfrequently encountered in sensors and instrumentation applications. The method is effective in jointly
estimating and tracking the signal amplitude along with additive DC bias, interfering sinusoids, or exponentially decaying transients. Lyapunov method is used to study the stability and convergence of the proposed technique. In addition, the extension
of the algorithm for multiple sinusoid case is also presented. Computer simulations and the experimental results are included
to demonstrate the effectiveness and applicability of the proposed scheme in applications like resolver, synchros, and power
system relays. It can easily be implemented in cost effective manner requiring minimal computation effort and hardware.
2007 Elsevier GmbH. All rights reserved.
Keywords: Amplitude demodulation; Amplitude estimation and tracking; Disturbance cancellation; Adaptive algorithms

1. Introduction
Amplitude demodulation nds many practical applications in elds like communication systems [1], sensor
instrumentation [24], signal processing, and system identication. Different approaches [1] are found in literature for
demodulation of amplitude modulated (AM) ordinary, single side band (SSB), and double side band (DSB) signals.
The lter method [1] is commonly used when the delay introduced by the ltering is tolerable. On the contrary, when
such delay becomes detrimental, synchronous techniques
are employed [2,3,8,9]. The technique in [2] is faster in
settling time when compared to the lter method, which,
however, is at the cost of precise clock timing relations. It is
also more susceptible to noise and systematic disturbances
at the input signal resulting in substantial distortion of the
demodulated signal. The closed loop techniques in [3] is,
Tel.: +91 080 25082316; fax: +91 080 25082321.

E-mail address: santanu@isac.gov.in.

1434-8411/$ - see front matter 2007 Elsevier GmbH. All rights reserved.
doi:10.1016/j.aeue.2007.06.006

however, slower than the techniques presented in [2], but

it has a better noise suppression capability. In [4] a visionbased navigational sensor (VISNAV) is presented that demodulates the multiple position dependent signals from the
position sensitive diodes (PSD) to obtain their magnitudes.
The calculated magnitudes are further used in the attitude
estimation algorithm for navigation solution. However, it is
observed that the complete system was constrained by the
computation complexity of amplitude demodulation, which
could only be tackled using multirate processing techniques
in a high performance DSP. Another interesting example
is presented in [5] where a phase-locked loop (PLL) based
real-time approach is proposed to measure particle velocity
using laser Doppler velocimeter data. The electrical signal
from the photo-detector has time varying amplitude modulated over a carrier and DC bias [4]. The analog PLL will
cause degraded tracking and estimation performance of the
particle velocity, as the DC bias and the time varying amplitude severely affects the stability and the range of operation
of the analog PLL [12]. However, with adaptive cancellation

S. Sarma / Int. J. Electron. Commun. (AE) 62 (2008) 438 444

of such disturbances and using the proposed technique as

an automatic gain tuning (AGC) circuit, performance of the
laser Doppler velocimetry can be improved. Besides, in application like relays that protects the power systems, the input modulated signal can also have a DC bias [6], or an
exponentially decaying transient [7], which severely effects
the estimation of the relay parameters for protecting the
transmission lines. To solve this, computationally intensive
DFT-based techniques are used to remove the additive DC
bias and the exponential transient from the modulated input
sinusoid signal. Moreover, these methods that are generally
ofine, do not work when even order harmonics are present
at the input signal [6]. Further, it has been observed in many
situations, such disturbances are time varying, and the techniques having the capability to adapt to such changes are
more effective.
In this paper, a simple computationally efcient, online
technique is presented that performs synchronous amplitude
demodulation, with the capability to adaptively cancel systematic disturbances like DC bias, decaying transient, and
low frequency interfering sinusoids. The method is effective in estimating and tracking the magnitude of an additive
DC bias, interfering sinusoid, and exponentially decaying
transient signal. The stability and convergence of the proposed algorithm is studied using Lyapunov methods and it is
showed that the proposed demodulator is globally exponentially convergent. Further, an extension of the algorithm is
also provided for multiple sinusoids. Computer simulation
and experimental results are included to illustrate the effectiveness and applicability of the method in many practical
scenarios.
The paper is organized in six sections. Section 2 describes
the structure of the algorithm and the basic assumptions.
The stability and convergence study is included in Section 3;
whereas extension of the algorithm for multiple sinusoidal
cases is included in Section 4. Section 5 includes the simulation result and different case studies. Finally, the conclusion
is drawn in Section 6.

x (t)

439

g0 s

m0

d(t )

x(t)

e(t)

2g1 s

Cos(1(t))

m1

Fig. 1. Block diagram representation of the algorithm.

with a DC bias component given as:

x(t) = m0 + m1 cos(1 (t)),

(1)

where m0 is the DC bias/disturbance, m1 and 1 (t) is the

magnitude and the phase of the sinusoid respectively. It is
assumed that the phase (or the frequency) 1 (t) = 1 t is
known and the disturbances and the magnitude of the sinusoid are constants. The problem considered here is to nd
the estimates of the amplitude and the disturbance in the
sinusoid, which is represented by m
0 and m
1 , respectively.
The block diagram of the algorithm is shown in Fig. 1
where an estimate of the incoming signal x(t)
given by
1 (t) cos(1 (t))
x(t)
=m
0 (t) + m

(2)

is produced and subtracted to obtain a difference signal

d(t) = x(t) x(t).

The difference signal is multiplied with

a sinusoid of the same phase and integrated to obtain magnitude of the sinusoid. Similarly, the integration of the difference signal gives the estimate of the DC bias. Thus the
equations describing the algorithm are as follows:
d
m
0 (t) = g0 d(t),
dt
d
m
1 (t) = 2g1 d(t) cos(1 (t)).
dt

(3)
(4)

3. Analysis
2. Adaptive synchronous amplitude
demodulator
The proposed algorithm is similar to least mean square
(LMS) adaptive algorithm which takes the advantage of
gradient techniques. Adaptive techniques have been classically used in applications where stationary and time invariant systems are assumed and the convergence properties
have been veried for a wide range of applications. However, they have also been used for non-stationary and time
varying systems with appreciable performance. The proposed demodulator has these advantages making it widely
applicable.
The block diagram representation of the proposed demodulator is shown in Fig. 1. Let us consider a sinusoid signal

The analysis of the proposed demodulator can be carried

out using similar approach as that of the LMS algorithm.
It can be shown to have unbiased global convergence by
using Lyapunovs method [11]. For this we consider the
error dynamics representation of the algorithm as described
in (8). When difference signal d(t) is expanded, we get
d(t) = (m0 m
0 (t)) + (m1 m
1 (t)) cos(1 (t)),

(5)

d
0 (t))
m
0 (t) = g0 (m0 m
dt
+ g0 (m1 m
1 (t)) cos(1 (t)),

(6)

d
m
1 (t) = 2g1 (m0 m
0 (t)) cos(1 (t))
dt
1 (t))cos2 (1 (t)).
+ 2g1 (m1 m

(7)

440

S. Sarma / Int. J. Electron. Commun. (AE) 62 (2008) 438 444

Dening the errors in DC bias and magnitude as m

0 (t) =
m0 m
0 (t) and m
1 (t) = m1 m
1 (t), we get

Condition (13) results in the derivative of the Lyapunov

function such that

0 (t) = g0 m
m
0 (t) g0 cos(1 (t))m
1 (t),

1 ) = 2g1 [m
0 +m
1 cos 1 (t)]2 < 0,
V (m
0, m

1 (t) = 2g1 cos(1 (t))m

m
0 (t) 2g1 cos2 (1 (t))m
1 (t).

which is, indeed, negative denite for all 1 (t). It is evident

that the system (8) is globally asymptotically stable if (13)
is satised, for all values of 1 . In other words, the estimator
described by (3) and (4) converges globally when both the
gains are positive. However, the convergence time of the
estimator is related to the initial condition of the states as it is
a closed loop parameter estimator. In similar approach, it can
be shown that the estimator converges for additive sinusoidal
disturbance and exponentially decaying transients.

(8)

Rewriting (8) and in state space form X(t)=A(t)X(t),

where
T
X(t) = [m
0 m
1 ] , we get

A(t) =

g0

g0 cos(1 (t))

2g1 cos(1 (t)) 2g1 cos2 (1 (t))


.

(9)

Thus the system is linear and time varying. The equilibrium

point of this error system can be found by setting the righthand sides of (8) to zero. It can be noted that the error
system (8) has only one equilibrium point at (0,0) for any
value of 1 (t). In other words, the system described by (3)
and (4) has innite equilibrium points and the estimator
provides an unbiased estimate of the magnitudes, which is
very important for the estimation of the DC bias and the
magnitude of the sinusoid to converge to the actual values.
The characteristic equation at a particular value of 1 (t) is
given by
|(sI A)| = s[s + (g0 + 2g1 cos2 1 )],

(10)

where I represent an unity matrix of order 2, |.| the determinant, and s the Laplace operator. From (10) it is evident
that this type 1 system is stable for positive values of g0 and
g1 that makes the root negative.
To determine the global convergence, we nd the stability
of the error system (8) using Lyapunovs method. Let us
dene a Lyapunov function [11] give by
V (m
0, m
1) =

m
20
m
2
+ 1,
2
2

4. Extension of the algorithm

The basic block diagram presented in Fig. 1 can be extended for multiple sinusoids with known frequencies as illustrated in Fig. 2. It can be noted that a special case of
the above extension is the case of periodic signals with harmonic contents, in which the extended portions to the basic
building block can be obtained by suitably multiplying the
phase component 1 of the cosine carrier. This simplies
the knowledge of the each frequency components and can
be obtained by using techniques like phase locked loop. In
other words, the proposed extension to the basic algorithm
can work like an online Fourier series analyzer and nd applications in many signal processing applications like speech
tone detection [13]. The extended version of the algorithm,
the output of each extended portions are added to produce
reconstructed signal. The difference between the input and
the reconstructed signal is used as the error signal for the
extended portions. The stability of the extended system can
also be arrived using similar method as in Section 3. Additionally, the estimation of the frequencies can be combined
with it making it a useful estimator and demodulator for FM
and PM signals.

(11)
x(t)

that satises the positive deniteness property [11]. We can

show the convergence of the system described by (3) and
(4) by showing the negative deniteness of the derivative
of the Lyapunov function. The derivative of the Lyapunov
function, given by
jV dm
0
1
jV dm
V (m
0, m
+
,
1) =
jm
0 dt
jm
1 dt

g0 s

x(t )

Cos(1(t))

2g1 s

2g2 s

Cos( n(t))

m1

m2

Cos(2 (t))

(12)

m0

d (t )

= [g0 m
0 + 2g1 m
1 cos 1 (t)]
[m
0 +m
1 cos 1 (t)]

(14)

2 gn s

mn

is negative denite i.e.V (m

0, m
1 ) < 0 for the condition,
g0 = 2g1 , g0 > 0, g1 > 0.

(13)

Fig. 2. Extension of the basic building block for multiple components.

S. Sarma / Int. J. Electron. Commun. (AE) 62 (2008) 438 444

Input Signal Tracking

Input Signal Tracking

3

xcap
x

Magnitude

Magnitude

0
-2

x1cap
x1

2
1
0
-1

10

10

20

30

Magnitude

Magnitude

m0cap
m0

0.5

60

70

80

90

100

m0
m0cap

1
0.5
0

10

10

Amplitude Tracking of the Desired Input Signal

20

30

40

50

60

70

80

90

100

Amplitude Tracking of the Desired Input Signal

m1
m1cap

Magnitude

Magnitude

50

1.5

1.5

1
0

40

Disturbance Signal Tracking

Disturbance Signal Tracking

441

0
-1
-2

10

time

Fig. 3. Magnitude estimation in presence of DC bias.

m1
m1cap

10

20

30

40

50

60

70

80

90

100

time

Fig. 4. Synchronous amplitude demodulator as used in resolver

and synchros.

5. Simulation results and case studies

Input Signal Tracking
Magnitude

-5
0

10

20

30

40

50

60

70

80

90

100

80

90

100

Disturbance Signal Tracking

Magnitude

0.02
0.01
0
-0.01
-0.02
-0.03

10

20

30

40

50

60

70

Amplitude Tracking of the Desired Input Signal

6
Magnitude

In this section, simulation results using Matlab-Simulink

for few practical scenarios are presented. Different interesting cases of the proposed algorithm is considered and simulation results showing the tracking of the desired input,
additive disturbance, and the estimation and tracking of the
magnitude of the input sinusoid is plotted.
The cases considered are practically encountered in many
elds as described in Section 1. The most common case is
shown in Fig. 3 that illustrates the estimation of the magnitude of the sinusoid, and the additive DC bias in presence
of noise. As seen from Fig. 3, the estimator gives an unbiased estimate and its convergence time can be reduced by
adjusting the gains.
Fig. 4 shows its use as an amplitude demodulator. Though
the demodulation of the SSB [10] modulated signal is performed with a DC bias of 1, the technique can demodulate
the ordinary amplitude modulated signal with modulation
index [10] greater than one, i.e. the case of over-modulated
signal. The ordinary AM signal has the form xAM (t) = [A +
m(t)] cos(c t) for which the modulation index is dened as
 = | min{m(t)}|/A [10]. It is interesting to note that even
though the estimator has not been developed for the ordinary
amplitude modulator, but it effectively demodulates the signal while adjusting the effect to the disturbance estimation
as seen in Fig. 5.

4
2
0
-2
0

10

20

30

40

50

60

70

80

90

100

time

Fig. 5. Ordinary amplitude demodulation with modulation index

greater than one (over modulation).

442

S. Sarma / Int. J. Electron. Commun. (AE) 62 (2008) 438 444

Input Signal Tracking

Input Signal Tracking

4
xcap
x

2
0

Magnitude

Magnitude

-2
-4

2
0
-2
-4

10

20

30

40

50

60

70

80

90

100

20

1
0

Magnitude

Magnitude

m0
m0cap

80

100

-1

1
0
m0
m0ap

-1
-2

10

20

30

40

50

60

70

80

90

100

20

40

60

80

100

Amplitude Tracking of the Desired Input Signal

Amplitude Tracking of the Desired Input Signal

3
m1
m1cap

2
0

Magnitude

4
Magnitude

60

-2

40

Disturbance Signal Tracking

Disturbance Signal Tracking

2
1

m1
m1cap

-2

-4
0

10

20

30

40

50

60

70

80

90

100

time

Fig. 6. Amplitude demodulation in presence of sinusoidal disturbance.

20

40

60

80

100

time

Fig. 8. Magnitude estimation in presence of sinusoidal disturbance.

Input Signal Tracking

Magnitude

6
xcap
x

4
2
0
-2
0

10

20

30

40

50

60

70

80

90

100

Disturbance Signal Tracking

Magnitude

6
m0cap
m0

4
2
0
0

10

20

30

40

50

60

70

80

90

100

Amplitude Tracking of the Desired Input Signal

Magnitude

4
2
0
m1
m1cap

-2
-4
0

10

20

30

40

50

60

70

80

90

100

time

Fig. 7. Amplitude demodulation in presence of exponentially decaying transients for relay applications.

The synchronous demodulation of the amplitude in presence of a sinusoidal disturbance is shown in Fig. 6. Both the
disturbance and the demodulated amplitude are of similar
magnitude and frequency.
One practical signal that is encounter in relays [6,7]
during short circuits is effectively tracked while estimating the transient involved in the input as shown in Fig. 7.
Many FFT-based techniques [7] which are computationally intensive are practically used to solve and estimate the
same.
Unlike the common amplitude demodulators, this demodulator does not introduce extra delay in the demodulated
signal, which is quite often an important requirement in
sensor processing and control applications. Such an example is the processing of the resolver amplitude modulated
signals in resolver-to-digital (R/D) converter [8,9]. In nonideal resolver due to the harmonic distortion, an additive DC
bias and harmonics contents is present in the output, which
degrades the performance of many R/D converters. In such
situations, the use of the proposed demodulator and its extension can signicantly improve the performance of the
R/D converters. In addition to that, it can also be used for
disturbance cancellations.
The amplitude estimation in presence of sinusoidal disturbance and exponentially decaying transients are shown in
Figs. 8 and 9, respectively. As seen for, the estimator has
work wells for both disturbance and magnitude decaying

S. Sarma / Int. J. Electron. Commun. (AE) 62 (2008) 438 444

Input Signal Tracking

Input Signal Tracking

Magnitude

10
xcap
x

5
0

10

20

30

40

50

Disturbance Signal Tracking

Magnitude

6
m0cap
m0

xcap
x

5
0
-5
-10

Magnitude

Magnitude

10

-5

443

10

15
20
25
30
35
Disturbance Signal Tracking

40

m0cap
m0

1
0
-1
-2
5

10

15

20

25

30

35

40

45

50

Amplitude Tracking of the First Input Signal Component

10

20

30

40

50

Amplitude Tracking of the Desired Input Signal

Magnitude

8
0

m1cap
m1

6
4
2
0

m1cap
m1

6
4
2
0

10

20

30

40

10

15

20

25

30

35

40

45

50

Amplitude Tracking of the Second Input Signal Component

50

time

Fig. 9. Magnitude estimation in presence of exponentially decaying

transients.

Magnitude

Magnitude

50

45

3.5
3

m2cap
m2

2.5
2
1.5
0

10

15

20

25

30

35

40

45

50

time

Fig. 10. Amplitude estimation for multiple components in presence

of sinusoidal disturbance.

exponentially. It is evident from the simulations that the

algorithm can nd multiple applications.
Its usefulness is also observed for amplitude-modulated
signals with multiple components as shown in Fig. 10. The
input signal consists of two components: one having constant amplitude sinusoid of frequencies 6 Hz and the other
a decaying exponential having frequencies of 5 Hz. A sinusoidal disturbance was also added to form the input signal.
The tracking of all the components are illustrated in Fig. 10
showing its broader area of applicability.
Further, experimental result obtained using an ADSP20262 kit for a modulated resolver output at 4 KHz is presented in Fig. 11, to validate the digital implementation of
the algorithm using bilinear transform. Both the input and
the output signals are suitably scaled to meet the ADC and
DAC ranges of the DSP board.

6. Conclusions
In this paper an adaptive technique has been proposed to
jointly estimate and track the amplitude of a sinusoid along
with the estimation and tracking of a DC bias, lightly damped
exponentially decaying transients, and low frequency sinusoidal disturbances. The algorithm is very simple and easy
to implement. The unbiased global convergence of the estimator has been shown using Lyapunov method. Its extension

Fig. 11. Demodulating resolver outpur at 4 KHz with ADSP-21062

Kit using the proposed method.

for multiple components has also been described. Different

cases of practical importance have been simulated showing
its usefulness and applicability. Experimental result of DSP
implementation using bilinear transforms in demodulating a
resolver sensor output at 4 kHz is also provided to show its
effectiveness.

444

S. Sarma / Int. J. Electron. Commun. (AE) 62 (2008) 438 444

References
[1] Yam Y-O, Wong K-H. Innovative demodulation method
for SSB technique. IEE Proc-Circuits Devices Syst
1999;146(3):14852.
[2] Koukourlis CS, Trigonidis VK, Sahalos JN. Differential
synchronous demodulation for small signal amplitude
estimation. IEEE Trans Instrum Measure 1993;42(5):92631.
[3] Talbot D.B. Closed loop amplitude demodulator, United
States Patent, Patent no. 4404194, September 6, 1983.
[4] Gunnam KK, Hughes DC, Junkins JL, Kehtarnavaz N. A
vision-based DSP embedded navigation sensor. IEEE Sensors
J 2002;2(5):42842.
[5] Duff AL, Plantier G, Valire J-C, Bosch T. Analog sensor
design proposal for laser doppler velocimetry. IEEE Sensors
J 2004;4(2):25761.
[6] Gu J-C, Yu S-L. Removal of DC offset in current and voltage
signals using a novel Fourier lter algorithm. IEEE Trans
Power Deliv 2000;15(1):739.
[7] Guo Y, Kezunovic M, Chen D. Simplied algorithms
for removal of the effect of exponentially decaying DCoffset on the Fourier algorithm. IEEE Trans Power Deliv
2003;18(3):7117.
[8] Bnte A, Beineke S. High-performance speed measurement
by suppression of systematic resolver and encoder errors.
IEEE Trans Ind Elect 2004;51(1):4953.
[9] Benammar M, Ben-Brahim L, Alhamadi MA. A novel
resolver-to-360 linearized converter. IEEE Sensors J 2004;
4(1):96101.
[10] Hsu HP. Theory and problems of analog and digital
communications. New York, NY: McGraw-Hill International
Editions; 1993.

[11] Sastry S, Bodson M. Adaptive control stability, convergence,

and robustness. Englewood Cliff, NJ: Prentice Hall; 1989.
[12] Stensby J. Phase-Locked Loops: theory and applications.
Boca Raton, Fla: CRC Press; 1997.
[13] Kilani MT, Chicharo JF. A constrained notch Fourier
transform. IEEE Trans Signal Process . 1995;43(2):205867.
S. Sarma received his Bachelor Degree
in Electrical Engineering from National
Institute of Technology, Agartala, (formerly, Tripura Engineering College) in
1999 and his M.Tech in Electronics and
Communication Engineering from Indian Institute of Technology Guwahati
(IITG) in 2002. Since 2002, he has been
with the Control System Group of ISRO
Satellite Centre as a researcher. He has
been involved in design and development of Attitude & Orbit
Control System, ASIC based design of Bus Management Unit for
different class of satellites, UML modeling of large software systems, and DSP based system design. His research interests include
modeling and analysis of control systems, embedded computing
systems, and VLSI signal processing. He has been the recipient of
the national talent scholarship, GATE scholarship, and the winner
of the gold medal of Tripura University. He is a member of IEEE
and IETE, India.