How to calculate the coefficients b in get_time_smoothed_representation

[guomin]

def get_time_smoothed_representation(
    spectral, samplerate, hop_length, time_constant_s=0.001
):
    t_frames = time_constant_s * samplerate / float(hop_length)
    # By default, this solves the equation for b:
    #   b**2  + (1 - b) / t_frames  - 2 = 0
    # which approximates the full-width half-max of the
    # squared frequency response of the IIR low-pass filt
    b = (np.sqrt(1 + 4 * t_frames**2) - 1) / (2 * t_frames**2)
    return filtfilt([b], [1, b - 1], spectral, axis=-1, padtype=None)

Didn't understand how to get b from the equation b**2 + (1 - b) / t_frames - 2 = 0