Code got simpler too:
github.com/aicodix/code...
@xdsopl.bsky.social
Author of Robot36 and Rattlegram
Code got simpler too:
github.com/aicodix/code...
Done! The new PAC coders now use a rank map instead of frozen bits. That means we can compute the map once and use it for every PAC(N, K) code without a frozen bit array! And just like with the 5G reliability sequence, we can simply filter out a smaller length from a larger RM rule construction. π₯°
02.02.2026 07:58 β π 0 π 0 π¬ 1 π 0Drawing the rest of the owl turned out to be much more convoluted than I imagined it π
01.02.2026 08:45 β π 0 π 0 π¬ 1 π 0Good morning. Let's do this!
01.02.2026 06:34 β π 0 π 0 π¬ 1 π 0Need to write this down before going to sleep: Because we can use the same RM rule construction for each PAC code rate, we only need to give the encoder / decoder the code and message length. That means no more handling of frozen bits, as we can have a reusable sequence under the hood. ππΎββ‘οΈπ
31.01.2026 21:40 β π 0 π 0 π¬ 1 π 0Must be that update I missed. Not connecting the washing machine to the internet was a good move.
29.01.2026 14:08 β π 1 π 0 π¬ 0 π 0Apropos smol: made a compressor for short English text.
github.com/xdsopl/smol
The world is smol π€£
26.01.2026 09:37 β π 1 π 0 π¬ 0 π 0Vocoder and Synth! π₯°
26.01.2026 08:41 β π 1 π 0 π¬ 1 π 0Looks like you are re-implementing GNU Radio. Way to go!
24.01.2026 12:05 β π 3 π 0 π¬ 1 π 024V COB LED lighting glued and soldered overhead, under the cupboard
Finally jumped the 24V COB LED lighting wagon from 12V halogen lamps. Soldering overhead from a weird angle is very interesting work .. but it had to be done. The old stuff is refusing to die (just repaired an electronic halogen lamp transformer) but I need more light in the kitchen.
22.01.2026 10:00 β π 0 π 0 π¬ 0 π 0cinnamon roll torn apart while still steaming hot
Turned out really good. π
18.01.2026 20:58 β π 0 π 0 π¬ 0 π 0cinnamon rolls baked at 160C for half an hour
And done! π
18.01.2026 20:55 β π 0 π 0 π¬ 1 π 0cinnamon rolls rose for six hours and are now baked
Couple of hours turn into several hours with sourdough instead of commercial yeast. No matter. Now we bake after 6 hours of rising.
18.01.2026 20:06 β π 3 π 0 π¬ 1 π 0cinnamon rolls rose for four hours
Four hours in, looking good. Let's give it another two hours. Yeah, we are going to bake at 9PM π
18.01.2026 18:31 β π 1 π 0 π¬ 1 π 0cinnamon rolls before rise
Sourdough cinnamon rolls, mark two! Very high hydration, barely able to handle. Now letting rise for a couple of hours in the warm oven. Fingers crossed! π€
18.01.2026 14:16 β π 5 π 0 π¬ 1 π 0heart shaped box of chocolates
Successive cancellation decoding is a lot like eating a box of chocolates: If you accidentally choose a very flavorful one somewhere in the middle of the consumption, the rest of the chocolates will taste dull.
15.01.2026 17:00 β π 0 π 0 π¬ 0 π 0RFnexus took my OFDM modem to the next level:
github.com/RFnexus/mode...
Hard to see the low-SNR region with a logarithmic scale, so here with a linear scale and a list size of L=128. We see that the normal Polar codes are getting lucky more often and the CRC oracle even helps the PAC codes under list decoding:
13.01.2026 07:35 β π 2 π 0 π¬ 0 π 0Yes, I am as dissatisfied as you are about that chart. Sorry.
12.01.2026 13:16 β π 2 π 0 π¬ 1 π 0A larger list size makes the list-decoded Polar codes with CRC behave better than the Polarization-Adjusted Convolutional codes in the low SNR regime. Both the PAC and BCH codes don't use an oracle here and I would need to run a simulation for weeks, just to show you the interesting QEF-region.
12.01.2026 13:05 β π 4 π 0 π¬ 1 π 0Health has returned. Wonder how long it is going to last ..
12.01.2026 12:16 β π 0 π 0 π¬ 0 π 0The honor is all mine! π
12.01.2026 10:45 β π 1 π 0 π¬ 0 π 0Exactly the reason why I made it all open source!
12.01.2026 10:16 β π 3 π 1 π¬ 1 π 0