Towards application-flexible embedded data acquisition : framework, compression, and synchronization

This work presents advances in the design and implementation of embedded data acquisition systems. Its contributions are organized into three interrelated domains. The first domain lies in Model-Based Engineering and introduces a modular, model-based firmware generation framework for heterogeneous hardware platforms. The framework abstracts platform-specific elements into a reusable library and employs formal timing diagram models to describe peripheral acquisition behavior. This enables automatic firmware generation that satisfies strict real-time constraints across diverse platforms. To demonstrate the framework’s added value, complex sampling schemes based on Compressive Sensing (CS) were integrated, allowing traditional Nyquist-rate sampling to be transformed into pseudo-random, a priori decimation schemes with configurable compression factors. The second domain focuses on Compressive Sensing. To overcome common limitations such as reliance on domain-specific sparsity assumptions, this work proposes data-driven reconstruction approaches using convolutional sparse coding and convolutional neural networks (CNNs). These methods significantly improve robustness and generalizability across application domains. Validation is provided through the reconstruction of bat echolocation scan patterns and cardiopulmonary signals, demonstrating effectiveness in bandwidth-limited and resource-constrained scenarios. The third domain addresses synchronization in multimodal and multi-node data acquisition systems. Synchronization challenges are tackled using a pseudo-random signal injection technique that enables sub-sample alignment across black-box sensor modalities. This approach is evaluated in a case study analyzing conservatory pianists during performance. Practical challenges arising from hardware tolerances, such as frequency offsets, are addressed through a solution that ensures synchronization stability, enabling reliable alignment of sEMG, motion capture, tissue oxygenation, and MIDI data. Bridging the second and third domains, CS is applied to the synchronization signal itself, and the impact of compression ratio on synchronization accuracy is systematically analyzed and generalized. Additionally, this work presents PhysioEdge, a multimodal, CS-enabled wearable platform designed for synchronized acquisition of physiological signals. The platform enables empirical evaluation of power-efficiency versus signal fidelity trade-offs in real-time health monitoring. Overall, this work provides a scalable and partly automated pathway toward synchronized, efficient, and application-agnostic embedded acquisition systems, with direct applicability in biomedical monitoring, musical performance analysis, and remote environmental sensing. This work presents a flexible embedded data-acquisition framework that can be applied across different applications. The framework combines modular system design with efficient data compression and accurate time synchronization, enabling reliable and efficient handling of sensor data on resource-constrained devices. The proposed approach is validated through practical implementations on embedded platforms. The results demonstrate that flexible, scalable, and application-independent data-acquisition systems are achievable, supporting both industrial and research use cases.