Can't reproduce BPLSD results in the paper

[inmzhang]

Hi,

I'm testing the performance of BPLSD decoder and try to reproduce the results in Fig.6 of this paper. I used the following configuration for the decoder:

SinterLsdDecoder(
    max_iter=30,
    bp_method="min_sum",
    ms_scaling_factor=0.625,
    schedule="parallel",
),

and the circuits built with https://github.com/gongaa/SlidingWindowDecoder. The results are as follow:

• bplsd is the decoder configured as above
• bposd0 is the order-0 BPOSD decoder
• bposd_order is the order-7 BPOSD decoder
• bplsd_r6_w6_c1_hx36 is the data from the paper retrieved from Zenodo

As expected, the results of bposd0 and bplsd are close. However, there is a gap between bplsd and bplsd_r6_w6_c1_hx36 and I don't know why.

The stats for bplsd_r6_w6_c1_hx36 looks like:

   1000000,       203,         0,   804.9,bplsd_r6_w6_c1_hx36,b9d8d085602084b73e97936e9e45bc8ba2023fb7909e097bedc9d286130df939,"{""code"":""BB_n72_k12"",""commit"":1,""d"":6,""decoder_args"":{""bp_method"":""ms"",""max_iter"":4,""ms_scaling_factor"":0.625,""schedule"":""parallel""},""p"":0.001,""rounds"":6,""window"":6,""z_basis"":""True""}",

it seems like using the BPLSD combined with OWD. But it should not outperform the ordinary BPLSD in my understanding.Did I miss any details that could account for the gap?

[timohillmann]

Hi, this is odd, could you post a minimum working example of your code?