Sparse Linear Prediction for Packet Loss Concealment in Networked Music Performances

We present a sparse linear prediction algorithm for real-time audio packet loss concealment. The method iteratively selects non-contiguous lags to model periodic signals using an Orthogonal Matching Pursuit (OMP)-based approach for lag selection. Implementation on ARM hardware achieves sub-millisecond processing time for model fitting and prediction times of a few tens of nanoseconds per sample, making the algorithm well-suited for networked music performance applications.