FlexCore at USENIX NSDI 2017
Watch FlexCore's video presentation at the 2017 NSDI symposium.
Complexity and throughput
Based on a novel Metric of Promise, MultiSphere estimates the probability of a node in the tree search to be part of the transmitted vector and then, depending on the number of available processing elements (PEs), partitions the Sphere Decoding tree search into sub-trees which preserve ML optimality and can be processed in a nearly concurrent manner.
Multisphere's single carrier latency and complexity
In single-carrier detection with 32 available processing elements, MultiSphere can reduce latency by more than an order of magnitude compared to state-of-the-art sequential Sphere Decoders and maintain a similar computational complexity. Additionally, in the high SNR regime, MultiSphere's latency with 32 processing elemenets is close to that of linear detection methods (i.e., the number of transmitting antennas).
(Complexity is measured in terms of Partial Distance calculations, Latency measured in terms of Visited Nodes - 8 x 8 MIMO, 10-16dB SNR, 16-QAM uncoded transmission, SD partitioning based on the exact transmission channel).