A spark ignited internal combustion engine is controlled in response to a self-learned TOB reference. The self-learned TOB reference is based on a difference between a learned TOB offset and a desired or target TOB, and a sensed TOB. The learned TOB offset at a given operating condition, such as charge pressure, can be found by interpolating between the learned charge pressure breakpoints in a TOB learning algorithm. The TOB learning algorithm can include using a filtered charge pressure value to indicate the engine load at which the TOB is learned. An index determination is made with a look up table with charge pressure as an input and an array index of learned charge pressure and learned TOB offset as outputs.