Improve `kmax` determination

[Dengda98]

Currently, in kmax formula, there's a static offset $\dfrac{5\pi}{\Delta h}$ , which may increase unnecessary calculation when the sampling frequency is low (e.g. at a large distance).

[Dengda98]

The tests show that this static offset is necessary for converge.

There's another reason for the low-frequency noise at a large distance.

[Dengda98]

fixed in #158 and #162 . kmax is not the problem, or say, itself a reasonable enough value. The problem is the low-freq noise when the time window length is large.