Alexios Balatsoukas-Stimming

Hernan Cordova, Alexios Balatsoukas-Stimming, Yunus Can Gültekin et al.

Min-Sum (MS) decoding is a popular low-complexity alternative to belief propagation (BP), retaining only the minimum incoming message magnitude during check-node (CN) processing, at the cost of systematic message magnitude overestimation. The scaled MS (SMS) decoder compensates for this effect using a fixed scaling factor. We propose the syndrome adaptive gain Min-Sum (SAGMS) decoder for quantum low-density parity-check (QLDPC) codes, which adapts the message gain online based on the fraction of unsatisfied stabilizers, requiring no per-code or per-noise level optimization. We show that the scaling factor required for SMS to match belief propagation decreases with the CN degree, so any fixed scaling optimized for one degree incurs into a growing penalty as the CN degree varies. SAGMS avoids this limitation by adapting the gain during decoding. Simulations on generalized bicycle QLDPC codes demonstrate that SAGMS matches or outperforms the frame error rate (FER) of an offline optimized SMS decoder. Moreover, SAGMS approaches BP performance and, under certain conditions outperforms it while retaining MS-level complexity.

Arwin Gansekoele, Alexios Balatsoukas-Stimming, Tom Brusse et al.

As telecommunication systems evolve to meet increasing demands, integrating deep neural networks (DNNs) has shown promise in enhancing performance. However, the trade-off between accuracy and flexibility remains challenging when replacing traditional receivers with DNNs. This paper introduces a novel probabilistic framework that allows a single DNN demapper to demap multiple QAM and APSK constellations simultaneously. We also demonstrate that our framework allows exploiting hierarchical relationships in families of constellations. The consequence is that we need fewer neural network outputs to encode the same function without an increase in Bit Error Rate (BER). Our simulation results confirm that our approach approaches the optimal demodulation error bound under an Additive White Gaussian Noise (AWGN) channel for multiple constellations. Thereby, we address multiple important issues in making DNNs flexible enough for practical use as receivers.

Freek Jochems, Alexios Balatsoukas-Stimming

Non-linear self-interference (SI) cancellation constitutes a fundamental problem in full-duplex communications, which is typically tackled using either polynomial models or neural networks. In this work, we explore the applicability of a recently proposed method based on low-rank tensor completion, called canonical system identification (CSID), to non-linear SI cancellation. Our results show that CSID is very effective in modeling and cancelling the non-linear SI signal and can have lower computational complexity than existing methods, albeit at the cost of increased memory requirements.

Andreas Toftegaard Kristensen, Robert Giterman, Alexios Balatsoukas-Stimming et al.

Neural networks have become indispensable for a wide range of applications, but they suffer from high computational- and memory-requirements, requiring optimizations from the algorithmic description of the network to the hardware implementation. Moreover, the high rate of innovation in machine learning makes it important that hardware implementations provide a high level of programmability to support current and future requirements of neural networks. In this work, we present a flexible hardware accelerator for neural networks, called Lupulus, supporting various methods for scheduling and mapping of operations onto the accelerator. Lupulus was implemented in a 28nm FD-SOI technology and demonstrates a peak performance of 380 GOPS/GHz with latencies of 21.4ms and 183.6ms for the convolutional layers of AlexNet and VGG-16, respectively.

Andreas Toftegaard Kristensen, Andreas Burg, Alexios Balatsoukas-Stimming

In this work, we use deep unfolding to view cascaded non-linear RF systems as model-based neural networks. This view enables the direct use of a wide range of neural network tools and optimizers to efficiently identify such cascaded models. We demonstrate the effectiveness of this approach through the example of digital self-interference cancellation in full-duplex communications where an IQ imbalance model and a non-linear PA model are cascaded in series. For a self-interference cancellation performance of approximately 44.5 dB, the number of model parameters can be reduced by 74% and the number of operations per sample can be reduced by 79% compared to an expanded linear-in-parameters polynomial model.

Jean-François Têtu, Louis-Charles Trudeau, Michiel Van Beirendonck et al.

Lyra2REv2 is a hashing algorithm that consists of a chain of individual hashing algorithms, and it is used as a proof-of-work function in several cryptocurrencies. The most crucial and exotic hashing algorithm in the Lyra2REv2 chain is a specific instance of the general Lyra2 algorithm. This work presents the first hardware implementation of the specific instance of Lyra2 that is used in Lyra2REv2. Several properties of the aforementioned algorithm are exploited in order to optimize the design. In addition, an FPGA-based hardware implementation of a standalone miner for Lyra2REv2 on a Xilinx Multi-Processor System on Chip is presented. The proposed Lyra2REv2 miner is shown to be significantly more energy efficient than both a GPU and a commercially available FPGA-based miner. Finally, we also explain how the simplified Lyra2 and Lyra2REv2 architectures can be modified with minimal effort to also support the recent Lyra2REv3 chained hashing algorithm.

Michiel Van Beirendonck, Louis-Charles Trudeau, Pascal Giard et al.

Lyra2REv2 is a hashing algorithm that consists of a chain of individual hashing algorithms and it is used as a proof-of-work function in several cryptocurrencies that aim to be ASIC-resistant. The most crucial hashing algorithm in the Lyra2REv2 chain is a specific instance of the general Lyra2 algorithm. In this work we present the first FPGA implementation of the aforementioned instance of Lyra2 and we explain how several properties of the algorithm can be exploited in order to optimize the design.