Advanced Analog Signal Processing (DFG)

Idea and aim


We propose to apply the concept of recurrent neural networks in analog signal processing, in order to fit the minimization problems required by vector channel equalization and iterative decoding of forward error correction codes.

Need for research


The need for reliable wireless transmission with high data rate is increasing continuously in an attempt to follow the strongly growing demand for wireless mobile services.
Because of the non-ideal physical channel between transmitters and receivers, and in case of MIMO transmission, the received signal suffers from intersymbol, interuser/intersubchannel interference.  To cope with this interference, and to improve the reliability of the digital transmission, cancellation techniques (vector equalization) and channel coding are applied. In many cases this results in high complexity (large number of operation per data symbol), especially in vector scenarios with high data rates.
Suboptimum soft-valued iterative schemes (“turbo” schemes) offer a good trade-off between complexity and performance, but for high data rates the digital signal processing power might be still too high with regard to practical applications.
One approach to cope with this problem is given by analog signal processing (ASP), since it possesses the potential of increasing the signal processing speed (desirable for high data rate transmission) and/or decreasing the power consumption (desirable for battery-operated mobile units and for “green communications”). This is of particular interest for the turbo schemes, because an exchange of soft values between equalizer and decoder is required, and soft values are a natural property of ASP.

Estimated findings and goals

The main target of this project is to get a deeper knowledge and information about the advantages of the analog signal processing approach, by merging expertise in analysis and simulations of iterative algorithms, and design/modeling of high-speed low-power analog circuits.

Tools and methods used

Methodology in this project relies on a strong, active, and bi-directional co-operation between two research groups, respectively focusing on (i) circuit modeling and design in Si/SiGe BiCMOS technology, and (ii) numerical and statistical analysis of suboptimum schemes for channel equalization and decoding.

Project participants

The project is a joint project of the groups of Prof. Schumacher from the Institute of Electron Devices and Circuits, and Prof. Lindner from the Institute of Information Technology. The project is founded by the "Deutsche Forschungs-Gemeinschaft (DFG)"(