Intern / Master Thesis (All Genders) - Machine Learning-based Channel Coding for Continous-valued Source Symbol Transmission

The Fraunhofer-Gesellschaft is one of the world’s leading applied research organizations. 76 institutes develop cutting-edge technologies for our economy and society - more precisely: 32 000 people working in technology, science, administration and IT. They know: If you join Fraunhofer, then you are able to promote change. For themselves, for us and for the markets of today and tomorrow. Channel coding and modulation for digital data, i.e. information represented as bitstreams, is a well-established field and replaced the earlier simple analog transmission schemes (e.g. Amplitude or Frequency Modulation). Modern channel codes such as LDPC and Polar codes already operate close to Shannon capacity and are widely used in practice. However, for continuous-valued source symbols equivalent coding schemes are still missing. Analog transmission has been explored in1,2,3, but these methods typically yield suboptimal performance, with a few exceptions in limited or impractical scenarios. **You are interested in combining research and practices and would like to develop further in the field of Machine Learning? Then have a look at our offer** **Here’s how you will make a difference** The position is offered under the **»Broadband and Broadcasting«** department. In the project that this thesis will be part of, we design machine learning based signal processing for physical layer communication. This thesis aims to design machine learning-based channel coding schemes for continuous-valued source data, drawing architectural inspiration from classical coding principles. In particular, the work will follow a model-based deep learning approach, similar in spirit to that proposed in4, where domain knowledge from communication theory is combined with the flexibility of neural networks. - **Explore the foundations**: You conduct a comprehensive literature review on generative models applied to physical layer communication. - **Shape innovative solutions**: You design and implement generative AI-based transmission schemes (e.g., using VAE, GAN, or diffusion models). - **Evaluate progress**: You evaluate the performance of these schemes against conventional digital baselines in terms of distortion, reliability, and efficiency. **What you bring to the table** - You **study**in the field of **communication theory**, **signal processing**, and **machine learning**: - You have a **solid understanding** of **physical layer concepts**, **including modulation** and **channel coding**. - You have a **hands-on** **experience**with **Python**and **machine learning frameworks** such as PyTorch or TensorFlow, NumPy, SciPy. **What you can expect** - **Organize your schedule**: Benefit from flexible working hours that are perfectly compatible with your studies. - **Become part of a creative team**: Experience an open and friendly working atmosphere in which your ideas are valued. - **Variety that inspires**: Look forward to divers tasks that inspire and challenge you. - **Innovation that inspires**: Exciting and pioneering projects that make a real difference. We will agree your start date and weekly working hours with you individually (for an internship at least three months). You can reduce your hours before exams and increase them during semester breaks. You can set your working days flexibly. After your studies, there are attractive opportunities to join the institute on a full-time or part-time basis. You can flexibly determine the working days of your fixed-term employment contract. Location: Erlangen Fraunhofer-Institute for Integrated Circuits IIS [1] M. Gastpar, B. Rimoldi and M. Vetterli, "To code or not to code," 2000 IEEE International Symposium on Information Theory (Cat. No.00CH37060), Sorrento, Italy, 2000, pp. 236 [2] P. A. Floor and T. A. Ramstad, "Shannon-Kotel’nikov Mappings for Analog Point-to-Point Communications," in IEEE Transactions on Information Theory, vol. 70, no. 4, pp. 2491-2517, April 2024 [3] Z. Xuan and K. Narayanan, "Low-Delay Analog Joint Source-Channel Coding With Deep Learning," in IEEE Transactions on Communications, vol. 71, no. 1, pp. 40-51, Jan. 2023, [4] N. Shlezinger, J. Whang, Y. C. Eldar and A. G. Dimakis, "Model-Based Deep Learning: Key Approaches and Design Guidelines," 2021 IEEE Data Science and Learning Workshop (DSLW), Toronto, ON, Canada, 2021, pp. 1-6,