Recursive kernels for time-frequency signal representations

Time-frequency distribution kernels which satisfy the desirable time-frequency properties and simultaneously allow recursive implementations of the local autocorrelation and the ambiguity functions are computationally efficient and prove valuable for on-line processing. The authors introduce a class of recursive kernels which apply modified comb filters at different timelags. The generalized Hamming, Blackman, and half-sine kernels are members of this class. These kernels have well known low-pass filter characteristics, lead to computational invariance under the kernel extent, and compete in performance with existing nonrecursive t-f kernels.