The concepts behind FIR design are very well known. In reality, an FIR filter is known as an inner-product. It is so well defined that efficient methods are very well known -- CIC and half-band. Further, it isn't even difficult to turn a least-squares "reduced basis" into a filter. Several "window functions" exists for those who don't yet fully understand how basic an FIR filter is. Further, "firpm" is very good at making an FIR filter.
that said, unless you already have a good "revolutionary" idea, I'd look at how even basic window and least squares come into play. The concepts will make you at least employable.
For example, windows -- the FFT is a *explicative* poor/basic transform. It takes N points in times, and _specifies_ n points in frequency. But that doesn't mean much to anything that isn't exactly one of those "n" points. So the window function determines how much you are willing to give up -- worse performance on the few exact frequencies you specify. In return, better performance on the other frequencies.
Because "inner-product" is the same as "FIR", your will find several common methods. Even in specific applications, it isn't hard to develop an inner product (FIR filter) that works.
That said, if you are very motivated, L1 minimization is now the new L2 minimization. And if you can explain L1 vs L2 vs L_inf, You'll certainly be employable. L1 norm minimization is variants on linear programming, which isn't nearly as easy as linear systems.
Further, you can choose a subject at random, and apply L1 regularization, or L1 minimization, and write a patent. (eg, from lunch to transportation to dating to family-issues to financing to etc...)