Estimating the Frequency of the Sinusoidal Signal using the
Parameterization based on the Delay Operators
Tung Nguyen Khac
a
, Sergey Vlasov
b
and Radda Iureva
c
Faculty of Control Systems and Robotics, ITMO University, Kronversky Pr. 49, St. Petersburg, 197101, Russia
Keywords:
Estimation Parameters, Identification Algorithms, Frequency, Sinusoidal Signal, Regressor.
Abstract:
The article presents an algorithm for estimating the frequency of an offset sinusoidal signal. Delay operators
are applied to the measured signal, and a linear regression model is constructed containing the measured
signals and the constant vector depending on unknown frequency. For the vector regression model, the method
cascade reduction is used. A reduction procedure is proposed that allows the original model to be reduced to a
reduced one containing a smaller number of unknown parameters. Finally, using the classical gradient method
was used to compare the efficiency of the proposed method.
1 INTRODUCTION
One of the main tasks in the design of automatic con-
trol systems is action alignment of parametrically in-
definite disturbing influences on the control object.
In the theory of linear systems, there is an internal
model principle for solving such problems. It is nec-
essary to build models of the reference and disturb-
ing influences. In the case of harmonic disturbances,
the model parameters will contain unknown frequen-
cies. The initial conditions will be set by unknown
displacement, amplitudes, and phases of the disturb-
ing signal harmonics. In this case, it is necessary to
apply adaptive internal models, which provide para-
metric identification possibility of the disturbing sig-
nal.
The task of estimating the parameters of sinu-
soidal signals is fundamental and, in addition to
theoretical significance, has wide practical applica-
tion (Stoica et al., 2000). Such problem can arise
during the synthesis of a compensation system for
a parametrically uncertain disturbance (Pyrkin et al.,
2015), for example, in precision displacement sys-
tems (Aphale et al., 2008).
One of the fundamental problems of control the-
ory is the problem of real-time frequency estimation
for a signal consisting of several sinusoids. The prob-
lem is studied in many branches of science: signal
a
https://orcid.org/0000-0001-6430-1927
b
https://orcid.org/0000-0002-8345-7553
c
https://orcid.org/0000-0002-8006-0980
processing, instrument making, adaptive control. The
problem of frequency estimation is widely presented
in practical applications, for example, in precision po-
sitioning systems in nanotechnology (Aphale et al.,
2008), in dynamic positioning systems for vessels ex-
posed to external disturbances such as waves, winds,
and currents (Yohei Takahashi et al., 2007), in power
systems for fault detection (Xia et al., 2012), (Phan
et al., 2016), etc.
As a rule, identifying unknown parameters is
posed from a set of measurements, estimating pa-
rameters in real-time using adaptive control, or com-
pensating for disturbances. The problem of identi-
fying harmonic signal constant frequency has been
well studied over the last decade, and a large number
of real-time algorithms have been developed. Many
approaches solve these problems. The most famous
is the least-squares method and its various modifica-
tions (Lijung.N, 1991). For real-time estimation, it-
erative forms of the least-squares method or gradient
integral algorithms can be used. In (Pyrkin A.A. and
S.A, 2015), an algorithm for continuous-time para-
metric estimation of all parameters of an indefinite
disturbance with a deterministic polyharmonic struc-
ture is presented. Standard gradient estimate is used
for identification. In (Vedyakova et al., 2020) algo-
rithm for estimating an asymmetric exponentially de-
caying sinusoid is considered. This problem is a spe-
cial case of the issue considered in this work in the
case of one harmonic in the spectrum of the signal un-
der study. The algorithm is based on the dynamic ex-
pansion of the regressor. In (Aranovskiy et al., 2016),
656
Khac, T., Vlasov, S. and Iureva, R.
Estimating the Frequency of the Sinusoidal Signal using the Parameterization based on the Delay Operators.
DOI: 10.5220/0010536506560660
In Proceedings of the 18th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2021), pages 656-660
ISBN: 978-989-758-522-7
Copyright
c
2021 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
algorithm of frequencies estimation of unbiased poly-
harmonic signal is presented. The algorithm is based
on the use of dynamic expansion of the regressor and
standard gradient estimation and provides exponential
convergence to zero of the estimation error. In (Ara-
novskii, 2008), algorithm of unknown frequency esti-
mation in continuous time of a displaced sinusoidal
signal is considered. The algorithm has noise im-
munity to amplitude-limited measurement noises and
provides asymptotic convergence to zero of the esti-
mation error.
In this article the synthesis of devices for signal
frequency estimation is considered. Parametrization
is proposed to obtain a linear regression model, the
vector of unknown parameters associated with the
original signal parameters. The cascade reduction
method is used to estimate the parameters (Bobtsov
et al., 2010), (Iureva et al., 2020). Conditions are for-
mulated under which the exponential convergence to
zero of the estimation errors is ensured.
This paper is organized as follows: problem state-
ment is described in Section 2; linear regression
model is constructed in Section 3; in Section 4 the es-
timation algorithm is proposed, and exponential con-
vergence of the estimation error to zero is proved; in
Section 5 proposed algorithm computer simulation re-
sults are included confirming the efficiency of the ap-
proach and finally the conclusion.
2 PROBLEM FORMULATION
Consider the measured offset sinusoidal signal:
y(t) = σ + νsin(ωt + ϕ), (1)
where ω R
+
is signal frequency, ν R
+
stationary
amplitude, σ R is the bias, ϕ is rare phase
and the number of signal harmonics y(t). Parameters:
ω, σ, ν, ϕ are considered unknown.
It is required to form estimations
ˆ
ω(t) of the fre-
quencies that ensure the convergence of the estima-
tion error of
e
ω(t) = ω
ˆ
ω(t) to zero under the fol-
lowing assumptions:
Assumption 1: Signal consists of one harmonic offset.
Assumption 2: Minimum frequency ω and maximum
frequency
¯
ω are known.
3 PARAMETRIZATION
Consider the measurable harmonic signal (1) with ex-
ponentially damped amplitude and bias. On the first
step the goal is to find linear regression model with
measurable variables and constant parameter associ-
ated with an unknown frequency ω.
Consider two signals:
y
1
(t) = y(t λ),t λ, (2)
y
2
(t) = y(t 2λ),t 2λ. (3)
where λ R
+
is chosen delay value.
Remark 1: The delay value λ from (2) and (3) should
be chosen such that λ <
π
¯
ω
.
The output signals (2) and (3) can be rewritten ex-
plicitly:
y
1
(t) = σ + a
1
νsin(ωt + ϕ) b
1
νcos(ωt + ϕ), (4)
y
2
(t) = σ + a
2
νsin(ωt + ϕ) b
2
νcos(ωt + ϕ), (5)
where a
1
= cosωλ, b
1
= sinωλ, a
2
= cos2ωλ, b
2
=
sin2ωλ and remark that a
2
= 2a
2
1
1, b
2
= 2a
1
b
1
.
Subtract from (1) multiplied by a
1
equation (4)
and subtract from (1) multiplied by 2a
2
1
1 equation
(5). Then is obtain:
a
1
y(t) y
1
(t) = (a
1
1)σ + b
1
νcos(ωt + ϕ) (6)
y(t)(2a
1
2
1) y
2
(t) =
= (2a
1
2
2)σ + 2a
1
b
1
νcos(ωt + ϕ),
(7)
Then subtract from (6) multiplied by 2a
1
(7) and ob-
tain:
2a
1
(a
1
y(t) y
1
(t)) (2a
1
2
1)y(t) + y
2
(t) =
= σ(2a
1
+ 2), (8)
y
2
(t) + y(t) = 2(a
1
1)σ + 2a
1
y
1
(t). (9)
Equation (9) can be written in the form of a linear
regressor with respect to two parameters a
1
, σ:
ψ(t) = ξ
T
(t)Θ, (10)
where
ψ(t) = y(t) + y
2
(t), (11)
ξ
T
=
y
1
(t) 1
, (12)
Θ =
2a
1
2(a
1
1)σ
=
θ
1
θ
2
. (13)
4 PARAMETER ESTIMATION
In the previous section linear regression model (10)
was gained. In this section the method for param-
eter estimation is proposed. Estimation algorithm
is presented based on the cascade reduction method
(Bobtsov et al., 2010).
Estimating the Frequency of the Sinusoidal Signal using the Parameterization based on the Delay Operators
657
Figure 1: Block diagram of the algorithm (18).
In this case we sequentially integrate equation (9),
i.e.:
Z
t
0
(y
2
(τ) + y (τ)) dτ =
= θ
1
Z
t
0
y
1
(τ)dτ + θ
2
Z
t
0
dτ, (14)
Introduce the notation:
α
1
(t) =
Z
t
0
(y
2
(τ) + y(τ))dτ,
α
2
(t) =
Z
t
0
dτ,
α
3
(t) =
Z
t
0
y
1
(τ)dτ.
and sequentially first divide by α
2
(t), and then differ-
entiate the last relation. Then get:
˙
α
1
α
1
2
α
1
˙
α
2
α
2
2
=
= θ
1
(
˙
α
3
α
1
2
α
3
˙
α
2
α
2
2
). (15)
Divide (15) into two parts by α
2
2
, and obtain:
˙
α
1
α
2
α
1
˙
α
2
= θ
1
(
˙
α
3
α
2
α
3
˙
α
2
). (16)
Introduce the following notation: β
1
=
˙
α
1
α
2
α
1
˙
α
2
, β
2
=
˙
α
3
α
2
α
3
˙
α
2
.
Then equation (16) takes the next form:
β
1
(t) = θ
1
β
2
(t). (17)
whence follows an identification algorithm in the
form:
˙
ˆ
θ
1
(t) = γ
1
θ
1
(t)β
2
2
(t) + γ
1
β
2
(t)β
1
(t). (18)
where γ
1
R
+
is the chosen constant that provides ex-
ponential convergence of the estimation error to zero.
From (17) and (18) can be obtained the differential
equations for errors checking:
e
θ
1
(t) = θ
1
ˆ
θ
1
(t).
Frequency Estimation
It follows from (18) that:
ˆ
ω(t) =
1
λ
arccos
ˆ
θ
1
2
!
. (19)
Figure 2: Block diagram of the algorithm (22).
Since the function domain (19) is the subset of R,
it is necessary to put some restrictions on it
ˆ
θ
1
(t). Un-
der Assumption possible values of θ
1
satisfying the
inequality:
2cos
¯
ωλ θ
1
2cosωλ. (20)
Rewrite estimation for
ˆ
θ
1
, which would satisfy the
next equation:
˙
ˆ
θ
1
(t) = Pr(γ
1
θ
1
(t)β
2
2
(t) + γ
1
β
2
(t)β
1
(t)). (21)
The projection Pr() allows condition (20) to be sat-
isfied so that the estimation remains qualitatively the
same (P.A.Loannou, 2012).
5 NUMERICAL EXAMPLES
In this section the simulation results are presented.
These results illustrate the efficiency of proposed es-
timation algorithm. All simulations have been per-
formed in MATLAB Simulink.
Let us compare the proposed algorithm with other
identification. The gradient descent method was taken
as an example.
The device for estimating parameters based on
gradient descent has the form:
˙
ˆ
Θ = Kξ
ψ ξ
T
ˆ
Θ
, (22)
where K R
+
is the chosen constant that provides
exponential convergence of the estimation error to
zero. Different signal was taken to check the algo-
rithm operation. This signal belongs to earlier con-
sidered algorithm for two different harmonics: y(t) =
4 + 2sin(2t + 2) and y(t) = 5 + 2sin(4t + 1). Delay
statements are used with the following delay values:
Method Cascade Reduction: γ
1
= 1, γ
1
= 20 and λ =
0.1, 0.3, 0.5.
The simulation results are shown in figures 3, 4, 5,
6.
Method Gradient Descent: K = 0.1, K = 0.5 and λ =
0.1, 0.3, 0.5.
The simulation results are shown in figures 7, 8, 9,
10.
ICINCO 2021 - 18th International Conference on Informatics in Control, Automation and Robotics
658
Figure 3: Parameter estimation transients for signal y(t) =
4 + 2sin(2t + 2) at γ
1
= 1 (Method Cascade Reduction).
Figure 4: Parameter estimation transients for signal y(t) =
4 + 2sin(2t + 2) at γ
1
= 20 (Method Cascade Reduction).
Remark 2: There is an optimal value at which the
speed is maximum in the gradient method, and for the
cascade reduction method, show that with an increase
in the value, the convergence time is much faster, the
speed can be increased infinitely. At the same time,
looking at the diagrams, we can see that when chang-
ing the delay operator, the method of reduction is al-
most unchanged, but for the gradient method, the con-
vergence time increases quite a lot and the overshoot
increases.
For the case of unknown parameters, numerical
modeling was carried out, which illustrated that when
using the cascade reduction method, the oscillations
in the parameter estimates were significantly lower,
and the response time was much faster than when us-
ing the gradient descent method. For the slope reduc-
tion method in both cases, the temporary time to esti-
mate the signal parameters is 450 seconds, compared
with 2 second for the cascading method.
The simulation results show that when using the
cascade reduction algorithm, the parameter estimates
are significantly lower and the response time is much
Figure 5: Parameter estimation transients for signal y(t) =
5 + 2sin(4t + 1) at γ
1
= 1 (Method Cascade Reduction).
Figure 6: Parameter estimation transients for signal y(t) =
5 + 2sin(4t + 1) at γ
1
= 20 (Method Cascade Reduction).
Figure 7: Parameter estimation transients for signal y(t) =
4 + 2sin(2t + 2) at K = 0.1 (Method Gradient Descent).
faster than when using the gradient method, and there
is almost no overshoot when using the cascade reduc-
tion method. Thus, the cascade reduction method may
be preferable for use in practical problems.
6 CONCLUSIONS
In the article the problem harmonic signal parame-
ters definition is considered. New parameterization
method based on operator delay application to mea-
surable signal is applied to construct linear regression
model. Methods for producing estimates of the fre-
quency of a harmonic signal are presented, making it
possible to obtain estimates of the parameters at a pre-
determined time. Computer simulation has been car-
ried out to illustrate the performance, demonstrating
the parametric convergence of the algorithm variable
(19), (22) to the correct value. Obtained algorithms
Figure 8: Parameter estimation transients for signal y(t) =
4 + 2sin(2t + 2) at K = 0.5 (Method Gradient Descent).
Estimating the Frequency of the Sinusoidal Signal using the Parameterization based on the Delay Operators
659
Figure 9: Parameter estimation transients for signal y(t) =
5 + 2sin(4t + 1) at K = 0.1 (Method Gradient Descent).
Figure 10: Parameter estimation transients for signal y(t) =
5 + 2sin(4t + 1) at K = 0.5 (Method Gradient Descent).
are supposed to compensate vertical inertial accelera-
tions in estimating gravity anomalies on moving ob-
ject. Future investigations will be devoted to extend-
ing the methodology to the case of multisinusoidal
signal estimation.
REFERENCES
Aphale, S. S., Bhikkaji, B., and Moheimani, S. (2008).
Minimizing scanning errors in piezoelectric stack-
actuated nanopositioning platforms. Nanotechnology,
IEEE Transactions on, 7:79 – 90.
Aranovskii, S.V., B. A. K. A. (2008). Identification of fre-
quency of a shifted sinusoidal signal.
Aranovskiy, S., Bobtsov, A., Romeo, O., and Anton, P.
(2016). Improved transients in multiple frequencies
estimation via dynamic regressor extension and mix-
ing. IFAC-PapersOnLine, 49.
Bobtsov, A., Kolyubin, S., and Pyrkin, A. (2010). Com-
pensation of unknown multi-harmonic disturbances in
nonlinear plant with delayed control. Automation and
Remote Control, 71:2383–2394.
Iureva, R., Kremlev, A., Subbotin, V., Kolesnikova, D.,
and Andreev, Y. (2020). Digital twin technology for
pipeline inspection. pages 329–339.
Lijung.N (1991). Systems identification. theory for the user.
19(2):432.
P.A.Loannou, J. S. (2012). Robust adaptive control. page
821.
Phan, A. T., Hermann, G., and Wira, P. (2016). A new state-
space for unbalanced three-phase systems: Applica-
tion to fundamental frequency tracking with kalman
filtering. pages 1–6.
Pyrkin, A., Bobtsov, A., Nikiforov, V., Kolyubin, S.,
Vedyakov, A., Borisov, O., and Gromov, V. (2015).
Compensation of polyharmonic disturbance of state
and output of a linear plant with delay in the control
channel. Automation and Remote Control, 76:2124–
2142.
Pyrkin A.A., Bobtsov A.A., V. A. and S.A, K. (2015). Ro-
bust and adaptive systems.
Stoica, P., Hongbin Li, and Jian Li (2000). Amplitude esti-
mation of sinusoidal signals: survey, new results, and
an application. IEEE Transactions on Signal Process-
ing, 48(2):338–352.
Vedyakova, A., Vedyakov, A., Bobtsov, A., and Pyrkin, A.
(2020). Drem-based parametric estimation of bias-
affected damped sinusoidal signals *. pages 214–219.
Xia, Y., Douglas, S. C., and Mandic, D. P. (2012). Adaptive
frequency estimation in smart grid applications: Ex-
ploiting noncircularity and widely linear adaptive esti-
mators. IEEE Signal Processing Magazine, 29(5):44–
54.
Yohei Takahashi, Shigeki Nakaura, and Mitsuji Sampei
(2007). Position control of surface vessel with un-
known disturbances. pages 1673–1680.
ICINCO 2021 - 18th International Conference on Informatics in Control, Automation and Robotics
660