MIMO detector implementations using high-level synthesis tools from different generations

Abstract

In 2011, we proposed a new receiver structure called Frequency domain linear MMSE filter with sphere detection for single-carrier frequency-division multiple access (SC-FDMA) uplink transmission. Frame error rate simulations and complexity estimations were used to define the most practical sphere detector algorithm for this receiver. High-level Synthesis (HLS) tools were used for comparing different architectures for the sphere detector. After 2011, the HLS design approach has gained more popularity and the tools have evolved. In this paper, HLS tools from different generations were used for implementing the same K-best list sphere detector. The results were compared and the overall experience of the optimization process and the evolution of the HLS tools was discussed. Additionally, the evolution of FPGA technology was discussed. In conclusion, the HLS tools have evolved into practical implementation tools even for high complex wireless communication algorithms. Moreover, 25% increase in throughput and 58% lower resource usage was achieved with the latest generation FPGA.

Configurable Quasi-Optimal Sphere Decoding for Scalable MIMO Communications

“…23. When compared with K-BEST (K = 8), chosen because of its similar detection performance for 64-QAM 4 × 4 MIMO [33], it offers almost identical performance but the resulting accelerator for R-BSFE requires 30% fewer LUT and 76% fewer DSP resources, as quoted in Table VI.…”

“…It enables quasi-ML accuracy and real-time detection [31] but is not configurable. K-best does permit configuration [32], allowing any number of symbols to be enumerated at each search tree level, but incurs costly sort operations in order to do so [33]. A K-best variant in [34] also mimics FSD in adopting a fixed tree structure for a given number of antennas and modulation scheme.…”

Base Station MIMO Detector Algorithm Implementations

“…In real-valued 4 × 4 64-QAM scenario this means there are eight layers and eight PED calculations on each layer. The receiver structure and sphere detector processing blocks considered in this paper are illustrated in Figure 2 [11]. In this paper, we consider two different processing blocks.…”

“…In addition 28 nm 20 nm Virtex 7 FPGA implementations [11] , the K-best LSD MIMO detector was implemented on the 28 nm and 20 nm Zynq FPGAs to get good understanding of the technology influence on the results. Table I describes the implementation results for maximum throughput designs.…”

“…In [10], we implemented the K-best LSD algorithm on a 40nm Xilinx Virtex 6 field-programmable gate array (FPGA) using an early version of Xilinx Vivado High-level Synthesis (HLS) tool to get a good understanding of its complexity. In [11], we implemented the same MIMO detector algorithm on 28 nm Virtex 7, 20 nm Virtex UltraScale and 16 nm Virtex UltraScale+ FPGA using the latest version Vivado HLS tool to evaluate both the evolution of HLS tools and the evolution of FPGAs. The HLS tools enabled the implementation and comparison of different macroarchitectures in a relatively short time.…”