Flexible N-Way MIMO Detector on GPU

Abstract

Abstract-This paper proposes a flexible Multiple-Input Multiple-Output (MIMO) detector on graphics processing units (GPU). MIMO detection is a key technology in broadband wireless system such as LTE, WiMAX, and 802.11n. Existing detectors either use costly sorting for better performance or sacrifice sorting for higher throughput. To achieve good performance with high thoughput, our detector runs multiple search passes in parallel, where each search pass detects the transmit stream with a different permuted detection order. We show that this flexible detector, including QR decomposition preprocessing, outperforms existing GPU MIMO detectors while maintaining good bit error rate (BER) performance. In addition, this detector can achieve different tradeoffs between throughput and accuracy by changing the number of parallel search passes.

LordCore: Energy-Efficient OpenCL-Programmable Software-Defined Radio Coprocessor

“…A LORD detector on nVidia Geforce FX 1700 graphics processing unit (GPU) is reported in [14]. MIMO detectors have been reported on Nvidia Tesla C1060 GPU in [15] and Nvidia Geforce 560 Ti GPU in [16]. These GPU chips have high power consumption implying that they would not be suitable for mobile handset implementations.…”

“…Both the algorithms were analyzed on the processor cores at 900-MHz clock rate. The LORD algorithm was executed with two layers: with two and four receive antennas and with QPSK, 16 Table III shows the results of these performance benchmarks. The LTE category four requirements (2 × 2 MIMO, 64-QAM modulation, 150 Mbit/s) can be achieved with a single core when using the MMSE algorithm or with three cores when using the LORD algorithm, which has better detection performance.…”

“…The SIC algorithm in [16] is assumed to have roughly the same detection accuracy than the MMSE algorithm, while the nML algorithm is assumed to have similar accuracy as the LORD algorithm. Table IV shows the performance comparison of MIMO detection algorithms, and the characteristics of the platforms executing these algorithms are listed in Table V.…”

“…A 2 × 2 MIMO detector in 16-QAM mode with one-way algorithm on Nvidia Geforce 560 Ti GPU in [16] has the same throughput as dual LordCore executing MMSE, but Lordcore has hundreds of times better energy efficiency. In 2 × 2 64-QAM, a single LordCore executing MMSE reaches 3× throughput compared to Geforce GPU with one-way algorithm, provides thousands of times better energy efficiency.…”

“…In 2 × 2 64-QAM, a single LordCore executing MMSE reaches 3× throughput compared to Geforce GPU with one-way algorithm, provides thousands of times better energy efficiency. On 2 × 2 16-QAM, a quad LordCore system executing LORD has a better performance than the two-way algorithm in [16], providing hundreds of times better energy efficiency.…”