Professional Documents
Culture Documents
INTRODUCTION
invariant models to describe their frequency translation properties [2]. Oscillators also constitute a
good example of systems whose noise performance
cannot be simply analysed using LTI models [3].
For the cases described above, LTV models may
be formulated to describe some or all of the aspects
of their small-signal behaviour. In general, it is not
easy to fully describe the behaviour of such models
due to the time-varying nature of its coefficients.
However, a number of theories have been proposed
for determining analytically the general solution of
arbitrary dynamical LTV systems [4, 5]. The theories presented in references [4] and [5] are based
on the definition of quantities called either essential D-eigenvalues or dynamic eigenvalues.
Although the terminology used in [4] and [5] differs greatly, many of the concepts expressed there
are related [6]. Given the similarities between
these concepts, it is preferred to use the unifying
term time-varying eigenvalues. Depending on the
mathematical framework considered for its definition, the solutions proposed for an arbitrary dynamical LTV system in terms of the time-varying
eigenvalues resemble those which can be formulated
for a dynamical LTI system in terms of the solutions
of the characteristic equation for a scalar dynamical
LTI system [4] or in terms of eigenpairs for systems
of LTI differential equations [5].
The aforementioned theories have been used in a
number of design paradigms. In [7], for instance,
the formulation presented in [4] was used to design filters with time-varying bandwidth which can
be used in control applications. The concepts presented in [5] were used in the assessment of the local
behaviour of a second-order oscillator implemented
using dynamic translinear synthesis techniques [8].
Time-varying eigenvalues may be also used to validate the local behaviour of rather uncommon systems such as the dynamic translinear circuit with
chaotic behaviour presented in [9].
Although it is clear from [4] and [5] how the timevarying eigenvalues must be computed, it is not
so evident how they should be interpreted unlike
the case of algebraic eigenvalues associated to LTI
systems. In this article, some arguments will be
presented in order to understand the meaning of
the time-varying eigenvalues and which kind of information they convey. The rest of this document
is organised as follows. In section 2, some funda-
system can be decomposed as a cascade of firstorder LTV systems. This property is also found
in LTI scalar systems. However, there is a substantial difference in the LTV case: given that in
general the composition of the operators present in
expression (4) is not commutative, the sequence in
which the first-order LTV systems should be connected in order to build the original n-th order system cannot be changed. In the LTI case this is
possible since quantities 1 (t), 2 (t), . . . , n (t) can
be chosen as constants and therefore they can be
permuted freely in expression (4).
According to classical results presented in [4,
11], if a set of n linearly independent solutions
y1 (t), y2 (t), . . . , yn (t) of equation (1) are known, it
is possible to calculate a set of time-varying eigenvalues 1 (t), 2 (t), . . . , n (t) as follows
i (t) =
i (t)
d
ln
dt i1 (t)
(5)
(6)
(7)
y (n) (t)
y1 (t)
y2 (t)
...
yi (t)
where y(t) is a function which depends on the time
y10 (t)
y20 (t)
...
yi0 (t)
..
.
.
.
.
varying coefficients. The problem of finding a set of
(i1)
(i1)
(i1)
time-varying eigenvalues 1 (t), 2 (t), . . . , n (t) asy1
(t) y2
(t) . . . yi
(t)
sociated to the LTV differential equation (1) can be
(8)
better understood if equation (1) is reformulated in Notice that i (t) is nothing more than the Wronterms of a differential operator Da of the form
skian of a set of i linearly independent solutions.
Given a set of i solutions of equation (1), it is possDa = n + an (t) n1 + . . . + a1 (t) + a0 (t) (2)
ible to formulate a new LTV differential equation
where and stand, respectively, for the operators of order i of the form
of differentiation with respect to t and scalar muly (i) (t) + bi (t)y (i1) (t) + . . .
tiplication, whereas a0 (t), a1 (t), . . . , an (t) are the
+ b1 (t)y 0 (t) + b0 (t)y(t) = 0 (9)
time-varying coefficients of equation (1). In expression (2), n stands for the n-th derivative of whose solutions are given by y (t), y (t), . . . , y (t)
1
2
i
a scalar function and it can be seen as the compos- provided that its time-varying coefficients
ition of n differential operators . Using operator b (t), b (t), . . . , b (t) satisfy the following set
0
1
n
Da , it is possible to express equation (1) as follows of linear algebraic equations
Da {y(t)} = 0
(3)
(10)
The differential operator Da of order n as given
T
T
where bi (t) = [b0 (t), bl (t), . . . , bi (t)] and vi (t) =
in expression (2) can be expressed as a composition
(i)
(i)
(i)
[y1 (t), y2 (t), . . . , yi (t)]. This statement will be
of n differential operators of first order as follows
used in the following discussion.
Da = [ n (t)] . . . [ 2 (t)] [ 1 (t)] (4)
3 A SIMPLE INTERPRETATION FOR
where 1 (t), 2 (t), . . . , n (t) are the time-varying
THE TIME-VARYING EIGENVALUES
eigenvalues associated to equation (1) and represents the composition operator. Intuitively speak- In this section, an interpretation for the timeing, equation (4) suggests that any n-th order LTV varying eigenvalues will be given. Although the ar-
guments which will be exposed here lack the necessary mathematical rigour to formulate a theorem
which will cover the most general case, it should
suffice to get an intuitive idea of the meaning of
these quantities. To simplify the discussion, expressions (6)-(8) will be applied in the computation
of the time-varying eigenvalues for the second-order
differential equation with constant coefficients
(22)
= Aem1 t + Bem2 t
= Cem1 t + Dem2 t
(25)
As it is shown in [12, 13], quantity b1 (t) defines essential properties of the stability of the solutions
of equation (23). For any set of linearly independent solutions y1 (t) and y2 (t) of equation (23), the
Wronskian is given by
R
y1 (t) y2 (t)
b1 (t)dt
det
=
e
(26)
y10 (t) y20 (t)
In the general case, if the scalar term accompanying the differential operator n1 in expression (2)
and the expanded form of the differential operator
given in expression (4) are compared, the following
relation arises
an (t) = (1 (t) + 2 (t) + . . . + n (t))
(27)
+ an y (n1) (t) + . . .
+ a1 y 0 (t) + a0 y(t) = 0
(28)
an (t)dt = T (F1 + F2 + . . . + Fn )
(29)
CONCLUSIONS