FPTC: A Fast Parallel Transform-based Codec for Efficient Asymmetric Signal Compression

Modern high-performance computing and IoT deployments generate massive signal data that must be compressed on low-power sensors and decompressed at scale on servers. Existing lossy compression methods rarely optimize both reconstruction quality and decompression throughput simultaneously. To address this, Ben Mechels and colleagues introduce FPTC (Fast Parallel Transform-based Codec), an asymmetric codec designed for this exact use case. FPTC pairs a lightweight sequential encoder that runs on resource-constrained devices with a massively parallel GPU decoder for server-side batch decompression. The encoder applies a windowed discrete cosine transform (DCT) to exploit frequency-domain sparsity, quantizes spectral coefficients using a hybrid three-zone mapping, and encodes the result with Huffman coding using a novel packing scheme. This pipeline is throughput-oriented on the GPU, maximizing performance without sacrificing reconstruction quality.

FPTC was evaluated on ten datasets spanning four signal domains: biomedical diagnostics, seismic reflections, power-grid production metrics, and meteorological recordings. Results show FPTC outperforms existing frameworks in compression ratio while maintaining competitive throughput, achieving multiplicative compression gains of 3.6x on power data, 3.1x on meteorological data, 1.5x on biomedical data, and 1.2x on seismic data over prior methods. The asymmetric design makes it particularly suitable for sensor networks and edge devices where encoder simplicity is critical, while leveraging the massive parallelism of GPUs for fast server-side decompression. This work demonstrates a practical path to efficient signal compression in heterogeneous computing environments, from IoT to HPC.