Signal processing and embedded intelligence

Signal processing is an area of science that is involved in the acquisition, representation, and manipulation of signals required in a wide range of practical applications. Our experience in this area is quite diverse, and the competencies include:

• virtual instruments based on advanced DSP technologies;
• program-controlled radio devices, including those based on non-uniform sampling;
• signal-dependent analysis of non-stationary signals, event-driven analog-to-digital conversions;
• biometrics and brain signal processing;
• use of biological feedback in medical rehabilitation;
• microminiaturization of data acquisition and processing systems;
• smart sensor and networked embedded system signal processing;
• applications of directional antenna arrays in wireless sensor networks;
• transistor-based UWB pulse generators and receivers;
• etc.

A part of our signal processing research consists of image processing in such areas as biomedicine (detection of melanomas), satellite data processing (analysis of the forests), intelligent transportation systems (detection of vehicles and pedestrians), security (recognition of people using images of the palm), agriculture (detection of crops and weeds), and more. For object detection, localization, and image classification tasks we research machine learning-based approaches, including deep convolutional and recurrent neural networks. The focus of our research deals with the acquisition of annotated training datasets by faster labeling methods and through the novel generation of synthetic data. Furthermore, we develop efficient video processing algorithms for use on devices with limited computation resources, for example, we have developed and implemented an object detection algorithm that can process up to 15 frames per second on a Raspberry Pi Zero platform.

In the embedded intelligence field we focus on computation and energy efficiency and develop novel architecture solutions for the implementation of data processing algorithms (transformed ANN architectures etc.) on FPGAs and heterogeneous SoC devices (in collaboration with TECNALIA, Infineon, BUT, etc. in H2020 3Ccar, Autodrive, ENACT, and other projects). This research results in enabling demanding perception processing (e.g. stereo-image processing and NN-based inference) on edge computing devices. Our expertise in embedded intelligence resulted in an award from Latvian Academy of Sciences for one of the most significant achievements in Latvian science in 2018 (work “An original approach for transforming the architecture of ANN into Field-programmable gate array”).

Our accumulated experience and expertise in signal processing and embedded intelligence has facilitated the development of wearable sensing fabrics, train integrity monitoring system, multi-modal FPGA-based biometric system, high-datarate wireless sensory acquisition nodes for industrial environments, hardware accelerators for image processing and depth sensing, vehicle-to-X communication systems, large-scale 100+ sensor network TestBed and many others. Currently, EDI experts work with hardware platforms ranging from the smallest microcontrollers to the most advanced state-of-art heterogeneous Systems-on-Chip and their expertise covers PCB design, low-level software, algorithm development, digital and analog circuit engineering, prototyping, and cooperative-system development.