Techniques are disclosed for performing branch metric computations/noise predictive calibration/adaptation for over-sampled Y samples. In one or more embodiments, the techniques employ a data processing apparatus (circuit) that includes a parallel to serial convertor configured to receive a first stream of sample data (e.g., Y samples) and a second stream of sample data (e.g., Z samples). The parallel to serial convertor is operable to combine the first stream of sample data and the second stream of sample data into a combined stream of sample data (e.g., combined Y and Z samples). The data processing apparatus (circuit) further includes a filter (e.g., a noise predictive finite impulse response (NPFIR) filter, a noise whitening filter, such as a noise predictive calibration/adaptation module (NPCAL) filter, and so forth) that is configured to receive the combined stream of sample data and whiten noise in the combined stream of sample data.