Photoplethysmography (PPG) signals are used in pulse oximetry to calculate peripheral capillary oxygen saturation (SpO2) levels. It is well known that the accuracy of the calculated SpO2 is susceptible to motion noise. Using field data collected from human subjects during rest and while exhibiting medium to high-intensity physical exercises, we show how sensor redundancy can mitigate motion effects by employing two channels of red and infrared PPG signals. We demonstrate the advantage of redundancy using two SpO2 calculation methods. These methods are widely used 'Red over Infrared' (RoI) method and the Discrete Saturation Transform (DST) algorithm both, assisted by pre-filtering by a heart-rate tuned comb filter. The data were collected from individuals at rest and while exercising. SpO2 levels are calculated for two channels, each using a red and an infrared PPG signal. We integrated the SpO2 computed from the two channels with a Kalman filter (KF) and computed the SpO2 mean absolute error (MAE) from each channel individually and after integrating the two channels. The SpO2 MAEs of each one of the two channels separately were greater (and statistically significant) than the SpO2 MAE obtained from the two channels integrated with a KF. We observed that when two channels of PPG (red and infrared) signals were available, then combining SpO2 levels obtained from these two channels exhibited a smoother and more reliable estimate of the blood oxygen saturation level as compared the SpO2 levels computed from each channel separately. The reason is that if the data for the two channels are collected at sufficient physical distance from one another, the motion artifacts affecting the two channels are uncorrelated. The KF is then able to use past measurements and modeling of the dynamics to attenuate the effect of the motion artifacts.