xf = (x[0] - b/a)*np.exp(-a*np.arange(len(x))) + b / a

is the initial value of the exponential decay.

To see why, let's first consider the case where b=0, so we have:

xf = x[0]*np.exp(-a*np.arange(len(x)))

In this case, we start with x[0] and each subsequent value is a fraction of the previous value, given by the exponential decay factor np.exp(-a). So the decay rate a determines how quickly the values decay to zero.

Now let's consider the more general case where b is nonzero. We can rewrite the equation as:

xf = (x[0] - b/a)*np.exp(-a*np.arange(len(x))) + b/a*np.exp(-a*np.arange(len(x)))

The second term is just like the previous equation, but with a constant offset of b/a. So we can interpret b/a as the initial value of the exponential decay, which is then shifted up by b/a.

This makes sense intuitively, since b represents a constant baseline that we want to subtract from the original signal x. So the exponential decay starts from x[0]-b/a, which is the initial value after the baseline subtraction.