Normalization of MTI Filters

I need to determine how various order MTI filters perform against a given clutter distribution. My only issue is, how to normalize the filter response [in frequency.] For example, a 2nd order MTI filter (coefficients 1 -2 1) has a gain in the middle of the passband of 6 (sum of weights squared). I compute clutter residue by multiplying the filter response by the clutter frequency spectrum (clutter is stationary so most of it is in the notch).

I want to know what the clutter level is (clutter improvement factor) after it passes through the MTI filter. I believe the correct normalization has the peak, in the center of the passband, at 6. However, I'm having trouble explaining why. It's possible the filter should be normalized so the value at the center of the passband is 1 (0dB). Can anyone provide justification for my answer OR explain why it's incorrect?

[Dean Ranmar]

Follow-up: I also have to analyze the effect of using an inverse MTI later in the processing. So, I need to know how to normalize the iMTI filter response. (As computed now, the gain at zero frequency, for a 3 pulse (order 2) MTI, is about 10^(4.5), which is clearly wrong.)

[Dean Ranmar]

BTW, if I use the above binomial coefficients (1 -2 1), the center passband peak is 4 (sum of absolute values of the weights) - at the output of the MATLAB fft algorithm.

I seem to remember the proper normalization has to result in an average gain of zero across the filter ....