$sinc(\pi(x-y))=S(x-y)=\sum_{n=0}^\infty \phi_n(x)\phi_n(y)$ where $\phi_n(t)$ is a Prolate Spheroidal Wave Function (PSWF). (You must account for the factor of $\pi$.)
Refer to "Sampling with prolate spheroidal wave functions", Walter and Shen.
https://www.researchgate.net/publication/268496813_Sampling_with_prolate_spheroidal_wave_functions
The PSWF, $\phi_n(t)$, is defined as the eigenfunctions of the convolution with $S(x-y)$ between $-\tau$ and $\tau$, and normalized under the real line. (See eqns. 1.2, 2.1 and 2.2).
The value of your $\lambda_i$ is unity for all $i$, due to the second half of eq. 2.2.
Section "2.2 Discrete Orthogonality" gives the derivation in the first three equations, ending at eq. 2.4.: $S(t)=\sum_{n=0}^\infty a_n(x)\phi_n(t)$. The value, $a_n$ is $\phi_n(0)$, yielding $S(x)=\sum_{n=0}^\infty \phi_n(0)\phi_n(x)$, and a shift by $y$, yields $S(x-y)=\sum_{n=0}^\infty \phi_n(x)\phi_n(y)$
ALTERNATIVELY, $\sin( \sigma(t-s))/(\pi (t-s)) = (\sigma /\pi) \sum_{n=0}^\infty (2n+1)j_n(\sigma t)j_n(\sigma s)$, where $\ j_n(x)$ is the spherical Bessel function.
Refer to "New method of computing prolate spheroidal wavefunctions and bandlimited extrapolation: A general approach to bandlimited Fredholm kernels" by V. Vaibhav. https://arxiv.org/abs/1804.04713