Enhanced Practical Integer-Forcing Source Coding
Rebekka Schulz, Sebastian Stern, Robert F. H. Fischer · 2025
This work considers one-shot integer-forcing source coding (IFSC), where the quantization and modulo reduction are done according to scaled integer lattices. This variant shows a loss in performance compared to IFSC employing high-dimensional lattices, which is caused by residual errors due to the modulo reduction. To overcome this problem, we propose an individual scaling of the symbols before quantization to reduce this overload error probability. We develop an algorithm to find the optimal preprocessing based on gradient descent. Additionally, numerical simulation results are presented to illustrate the theoretical findings. It is shown that already this simple scaling can achieve significant gains at only little additional cost in the calculation of the required matrices.