The least-squares ambiguity decorrelation adjustment: its performance on short GPS baselines and short observation spans