The model will be,

y(t)=a*y(t-1)+err(t)  

We can write it in a lag format

y=a*B*y+e   ---->   y(1-a*B)=e

If it can be inverted, then y=e/(1-a*B)  or y= e(t) +a*e(t-1)+a^2*e(t-2)++a^3*e(t-3) +...     or |a| <1

|a|<1 if y(t) is stationary or no unit root.

You may take a look of any time series book.

MLE deosn't depend upon distribution form

英文看不懂，希望大家写中文