skypitcher / hats

Hyper-accelerated tree search (HATS) algorithm for solving integer least-squares problems in large-scale systems.

Hyper-Accelerated Tree Search (HATS)

This is the repo for the paper entitled "Towards Optimally Efficient Search with Deep Learning for Large-Scale MIMO Systems". The key idea of this paper is to train the network towards the optimal heuristic. You can check the Preprint online if you are interested.

Requirements

To run the code you have to fullfill the following dependencies,

• matplotlib>=3.1
• tabulate>=0.8
• numpy>=1.16
• pytorch >= 1.5
• sortedcontainers>=2.3

Benchmark

You can change n_ant and snr_list and then run python test.py to see the result. Note that only n_ant in [4, 8, 12, 16, 20, 24, 28, 32] and snr_list in range(5, 26) are supported by the out-of-box, since we only train the model for these cases.

You may see some results like the following,

Testing n_ant=32 snr=13 packet=141/1000 timeslots=16/16 algorithms=7/7 HATS(inf)..
NAME         BER                             PRECISION  STEPS
-----------  ---------------------------  ------------  ------------------
MMSE         2.495429e-02(3603.0/144384)  6.925975e-06  0.00/0.00
SMA*(128)    2.077793e-05(3.0/144384)     6.925975e-06  22976.27/261657.90
HATS(128)    3.462987e-05(5.0/144384)     6.925975e-06  347.41/1233.47
SMA*(16384)  2.077793e-05(3.0/144384)     6.925975e-06  22976.27/61161.44
HATS(16384)  3.462987e-05(5.0/144384)     6.925975e-06  347.41/700.78
SMA*(inf)    2.077793e-05(3.0/144384)     6.925975e-06  22976.27/45952.54
HATS(inf)    3.462987e-05(5.0/144384)     6.925975e-06  347.41/700.78

Under Construction

TBD

License

Anti 996 License Version 1.0