Best candidate is not displayed

[DavidAU652]

OS version: Windows 11 Pro 25H2

GPU: RTX 4070

CUDA toolkit version:
NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2025 NVIDIA Corporation
Built on Wed_Jul_16_20:06:48_Pacific_Daylight_Time_2025
Cuda compilation tools, release 13.0, V13.0.48
Build cuda_13.0.r13.0/compiler.36260728_0

Unexpected behavior:
No best candidate or score plot is displayed.

Reproduction steps:
Record an DMR transmission.
Demodulate it in FSP.DMRCrack
Try to brute force it

Tried building from source with same results.

[felixx-sp]

Hey! Sorry for the delay, I'm available again. Let me take a look, thanks for the feedback.

[felixx-sp]

This should be fixed in v0.3.2, there were some issues with how the GPU kernel was being launched. Update and let me know if it still happens!

[DavidAU652]

Yeah, I can now see the best candidate, and the UI is much better than before.
The only issue I’ve noticed so far is that the best candidate stays as "ABAB ABAB AB" and doesn’t seem to change. I’ll let it run until completion and let you know whether it still remains "ABAB ABAB AB".

[felixx-sp]

Thank you so much for your support! I'm trying my best to improve the program and make it a valuable tool for the community. That key is the result with the highest current score from the priority scan (it first checks for common patterns before the full 40-bit scan). What score value is displayed next to it? If the score is lower than 7-8, it's just background noise, and the real key will replace it as the search progresses

[DavidAU652]

The score of that key is 2474.44. It's still the best one, but only 18% of keys was checked. I should also note that I've recorded the audio sample in SDRConsole without using squelch.

[felixx-sp]

Hi David, any news?

[DavidAU652]

It finished right now with "ABAB ABAB AB" and thats not the correct key. Could it be a problem with the sample?

[felixx-sp]

For reference, here's a few lines of how a correctly generated .bin looks like

82A4829C5FC44AF28ABF4F118347F7D5DD53DFD54D43BA99AB00107E58E4C7B285;ALG=21;KID=0F;MI=531699C4;SILENCE=1
B5983AD7874CE1B3FCC302D9DE5134100000091590BEA0A6878D4C695337D2D32B;ALG=21;KID=0F;MI=531699C4
BEF6697B85313CC465B6A4346DB1700094100748F2FAD3679E643C7CE56D7319F5;ALG=21;KID=0F;MI=531699C4
5B986F789A399F8CE4598A50AAB170003284874CF32C1F203621F927AA66C47963;ALG=21;KID=0F;MI=531699C4
29E7BC0FFE5EA2AEDCC126CF61D154222282207D25094B3B2A37FFBB32E8AF7326;ALG=21;KID=0F;MI=531699C4
81521AE22FB42C44A28CD4E52991104AB0857E2AD831C86181B5A7446FF895B8D4;ALG=21;KID=0F;MI=531699C4
17240E1A3078299372A0520E6F07F7D5DD57DFDCBA844EEA1C4AE51C58C716D4B2;ALG=21;KID=0F;MI=178D0B9B
5A69B2CB7D25985F190A0510AC41341000000912BB2182480EAD08FC73715DA0A9;ALG=21;KID=0F;MI=178D0B9B
239A257CEE1E3F146323B8B172B1700094100744071BB08D5C819E7F8574A77EB6;ALG=21;KID=0F;MI=178D0B9B
868C38E62CD1DE0208B952DACBA1700032848740090F77B4B8EDC68C954D232CAE;ALG=21;KID=0F;MI=178D0B9B
61D31DB077BC186DCAB8EF87EFF154222282207DC6E5E3A6A4176009ADFE972B84;ALG=21;KID=0F;MI=178D0B9B
D83CF6DC957CDED1A0EF2FEC7811104AB0857E28B326CC590577561AA4B668F8C5;ALG=21;KID=0F;MI=178D0B9B
3A1097C2F2391D82572518254F67F7D5DD57DFDFCCDDDCE3CFE33B9EF10DF71547;ALG=21;KID=0F;MI=7F900958
BA75ED275DBDEB990A755F1E99E134100000091C2B644515E8BDA9B586F3EEC6FB;ALG=21;KID=0F;MI=7F900958
00E6F97D68B9E73F70364CA37C41700094100746FD2E849D6DFB8D099B5FA9113D;ALG=21;KID=0F;MI=7F900958

mine has 126 lines, all with exactly 66 hex chars, ALG=21, KID=0F, and MI= on every single line (21 distinct MI values, grouped 6 lines per MI since each superframe covers 6 bursts).

If your file is missing MI values on many lines, or has inconsistent KIDs, or the hex before the ; isn't exactly 66 characters, that's the root cause. The app needs that metadata to use the reliable KMI9 scoring path. Without it, it falls back to a statistical mode that's much weaker and can latch onto false patterns like ABABABABAB

Can you paste the first 5–10 lines of your .bin here so I can compare?

[DavidAU652]

Here is a few lines. I think that I see the problem. That is this "SILENCE=1" its on all of the lines. I will get a new sample and try again.

A481B272492096800C082097B1E744FC7DE21F752E4AE0110A5425803429E03770;ALG=21;KID=2D;MI=82BF1DB0;SILENCE=1
FF040968DC8E109220CCA2CEEFAC3480A030A227A7B282109E2ED06BCB06B1B162;ALG=21;KID=2D;MI=82BF1DB0;SILENCE=1
43CC102A0842644A621B48A790BC70A0F4003C7E2321DBED2F1074AA0A8C9035A9;ALG=21;KID=2D;MI=82BF1DB0;SILENCE=1
EA620C765B781C08C2D43606202C701002742875778F0749E47B09B7FE85EDD1AF;ALG=21;KID=2D;MI=82BF1DB0;SILENCE=1
48FF2182C4341D1D8CDBC112941C548282B22B444424B89B487AA0CF4812EF496C;ALG=21;KID=2D;MI=82BF1DB0;SILENCE=1
8F1295DC000C6D20AC0C91B3EA2C10EE30CD341D330E19C20EA24572AB88C9D9C8;ALG=21;KID=2D;MI=82BF1DB0;SILENCE=1
30F136482CFC0D8A28B876BA28A211FD78B74B3230A1A4A1542B183DC3F1E92AC3;ALG=21;KID=2D;MI=A092909E;SILENCE=1
278A312E010F22F84B940738915C3480A030A222BC79E2D8EC902AD229EE082E44;ALG=21;KID=2D;MI=A092909E;SILENCE=1
448296CE69D5C18284F2CDCDDAE355F878B21E2BA3330A32AE0421C648070FA247;ALG=21;KID=2D;MI=A092909E;SILENCE=1
440E1DA23E723B972989F23BF2BE3080A030A22C48712CAB0C8841087F403A8874;ALG=21;KID=2D;MI=A092909E;SILENCE=1
BA0680C89D3C02FBC7BEFE406C4860A0F0403C753AA60F03EF0066C7B1C1826208;ALG=21;KID=2D;MI=A092909E;SILENCE=1
1DFB749074CEB44FA8A5C29D58FC705002702E7C2B30A478922C884A1219AF8144;ALG=21;KID=2D;MI=A092909E;SILENCE=1
14FD3E43A5A6322BB6C0C3CB7BCC548282B22B47AF32B2670A13EDB701B16CAAC5;ALG=21;KID=2D;MI=F37BDBD6;SILENCE=1
8A2C8A09D0A50C1602446E320CCC10EE30D8211EC0A686B11F704837F082BACB6B;ALG=21;KID=2D;MI=F37BDBD6;SILENCE=1
132EE926DE10EAE2A694F20B08C211ED7CB74B280B11B73C0F8034E7ABC211598B;ALG=21;KID=2D;MI=F37BDBD6;SILENCE=1
C757A4AAE2C0BBB6B86CB2A2CCFC3480A030A220BC32AB02BB40784E8A0EE18D00;ALG=21;KID=2D;MI=F37BDBD6;SILENCE=1
7BBD58B5328B2A210EEDA2AB60BC70A0F4403C7F6099320037F05863320AF6C470;ALG=21;KID=2D;MI=F37BDBD6;SILENCE=1
E69D521881F107B864A0A8B05F08605002742C72DD2E2D8F78C8B09846A0001AA2;ALG=21;KID=2D;MI=F37BDBD6;SILENCE=1
8F127AD88A81EA456102C83880FC548282B22B0BBE2E1810D6AFAC7CB3C2BE0E3F;ALG=21;KID=2D;MI=17FAC63B;SILENCE=1
42802930E800182FD184D63920AC10EE30DD210482E0E318601CACCE92A2B14095;ALG=21;KID=2D;MI=17FAC63B;SILENCE=1
322F9D60FA5A0AC22D4923C85A7755FC78B61F3B38936B7B53B9218E18330F8AC9;ALG=21;KID=2D;MI=17FAC63B;SILENCE=1
36289C76A0310B2422DC7D8168AC3080A020A22F6E51B0BE7E3D40C8929CB00603;ALG=21;KID=2D;MI=17FAC63B;SILENCE=1
0F930B821876E06080B8393A19EC70A0F4403C79AD1E1AD0328B8232E0065CE2CE;ALG=21;KID=2D;MI=17FAC63B;SILENCE=1
CE87382CE11F98E18A007203882C704002742C29CEBAB48BA963A1D3AA11807370;ALG=21;KID=2D;MI=17FAC63B;SILENCE=1
0B46DD0A175280AD13861C1F8DFC548282B22B4B80980BE30EF21A8B91D65E2C90;ALG=21;KID=2D;MI=1D2725B2;SILENCE=1

[felixx-sp]

Hi David. Did you get a good sample?

[felixx-sp]

Found the root cause — thanks for pasting the .bin, that was the key.

Every line in your file is tagged SILENCE=1. The cracker uses SILENCE-tagged frames as a known-plaintext fast-reject filter (they must decrypt to all-zero C0/C1), so it hard-rejects any key that fails. When real speech frames are wrongly tagged silence, the correct key gets rejected and you're left with a meaningless high-score pattern like ABAB ABAB AB.

The over-tagging comes from your capture's DSP structure (a marker before every voice burst trips the "start-of-transmission silence" heuristic). The reference .bin above only has SILENCE=1 on the first burst of each transmission.

Fixed for the next build: when an implausible share of frames is tagged silence, the tool treats the tagging as unreliable, skips that fast-reject filter, and falls back to full scoring — so the correct key can no longer be pruned. I'll ping you when it's out; a retest with your same sample would be great.