Singular Value Decomposition for Compression of Large-Scale Radio Frequency Signals

The paper proposes an efficient matrix factorization-based approach to large-scale radio frequency (RF) signal compression tasks. While data compression techniques can significantly reduce the storage requirements and memory bandwidths for many types of data including image and audio files, a reasonable implementation for RF signals is less explored. However, since recorded RF signals can be extremely large, they often significantly impact the storage and handling of the data. In this paper, we focus on software defined radios (SDR) that process RF signals in the in-phase (I) and quadrature (Q) time samples, which are then transformed into a time-frequency representation. We investigate the use cases of the singular value decomposition (SVD) algorithm, which reduces the dimension of the time-frequency representation of the IQ samples, forming a low-rank approximation of the original. We validate the proposed method in various lossy RF signal compression tasks that show fast and reliable compression results with acceptable reconstruction error.