C2_W4_Lab_1 Intuition behind Timestamp preprocessing

what is the intuition behind this?

# Convert timestamps into periodic signals
  outputs['Day sin'] = tf.math.sin(timestamp_s * (2 * pi / day))
  outputs['Day cos'] = tf.math.cos(timestamp_s * (2 * pi / day))
  outputs['Year sin'] = tf.math.sin(timestamp_s * (2 * pi / year))
  outputs['Year cos'] = tf.math.cos(timestamp_s * (2 * pi / year))

Please read attention is all you need this paper to understand what positional encoding (till end of section 3.5) means. This concept is covered in deep learning specialization in course 5.