I'm currently studying Volterra series as an input-ouput-representation of nonlinear systems. For this, I found a variety of interesting papers. For instance, Lesiak & Krener (1978), Brockett (1976) and Boyd & Chua (1985). But it seems like there are two approaches to the Definition/Derivation of Volterra series.

I like the approach in Lesiak & Krener the most, because there, the authors show their results on existence and uniqueness with mathematical rigour. For an analytic control-affine system $\dot x=f(x)+g(x)u$ there exists an Operator $\psi$ mapping input functions $u$ from a suitable input function space to Outputs $y$ contained in a suitable output function space. This Operator possesses the following form:

$y(t)=(\psi u)(t)=w_0(t)+\int_0^T w_1(t,\tau_1)u(\tau_1)d\tau_1+\int_0^T\int_0^{\tau_1}w_2(t,\tau_1,\tau_2)u(\tau_1)u(\tau_2)d\tau_1 d\tau_2+\ldots$

The second approach is done via the buzzword 'multiple convolution', i.e. for LTI Systems $\dot x=Ax+Bu$, $y=Cx$, $x(0)=x_0$ we have $y(t)=Ce^{At}x_0+\int_0^tCe^{A(t-\tau)}Bu(\tau)d\tau$. If we set $x_0=0$, then we obtain the standard convolution formula $y=w_1\ast u$, where $w_1=Ce^{A(\cdot-\tau)}$ corresponds to the first order Volterra kernel. Authors now extend this approach by adding (probably infinitely) many summands, thus rendering the multi-convolution formula:

$y(t)=(Nu)(t)=h_0+\int h_1(\tau_1)u(t-\tau_1)d\tau_1+\int h_2(\tau_1,\tau_2)u(t-\tau_1)u(t-\tau_2)d\tau_1d\tau_2+\ldots$

Q1: How do these two approaches correspond to each other? That is, how can I transform kernels $w_i(t,\tau_1,\ldots,\tau_i)$ and inputs $u(\tau_1),\ldots,u(\tau_i)$ to time-invariant kernels $h_i(\tau_1,\ldots,\tau_i)$ and inputs with retarded arguments $u(t-\tau_1),\ldots,u(t-\tau_i)$? In the case it does, how does the integration domain change/what does it look like for the operator $N$?

Q2: Being an approximation method that Volterra series are, how can I apply the Stone-Weierstrass-Theorem to a nonlinear system $\dot x=f(x,u)$, $y=h(x)$, if $f,g,h$ are analytic or $k$ times continuously differentiable? This means, how does the derivation look like, if one studies more general systems than control-affine systems as was done in Lesiak & Krener.

I greatly appreciate any help, whether it is a full proof/derivation argument, a helpful hint or another paper with good explanations!

Best regards, Joe


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.