Towards Practical Polar Codes With Low Scaling Exponent
summary
Construction uses generalized concatenation: permuted Arıkan inner kernels combined with Reed‑Solomon outer kernels and recursive ML kernel processing based on Fujiwara et al., trading partial distance profiles against punctured‑code dimension. Results show K=64 kernels with scaling exponent 3.13 and recursive processing cost about 10^5 summation/comparison ops, while a handcrafted K=16 kernel gives lower cost and better complexity–performance; simulated polar subcodes (n≈4000, SNR 1.25 dB) beat 5G LDPC.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 0:22Precoded polar codes overview
- 1:09Polarization kernels properties
- 2:56Practical kernel challenges
- 4:25Recursive ML decoding
- 7:21Polarization rate and distances
- 8:49Generalized concatenated codes
- 10:38Combining small kernels
- 13:37Kernel construction examples
- 15:56Row permutation techniques
- 17:26Comparison with state-of-the-art
- 19:15Polar subcodes performance
- 21:55Simulation results
- 23:00Comparison with 5G LDPC
- 24:09Conclusion
- 24:32Question and answer session
related talks
28mCommunication Avoiding Algorithms for Linear Algebra and Beyond
27mList Recovery and Local Decoding of Expander Codes
15mLESS is More for I/O-Efficient Repairs in Erasure-Coded Storage
17mParaRC: Embracing Sub-Packetization for Repair Parallelization in MSR-Coded Storage
57mCoding for Interactive Communication
44mTutorial on List Decodable Coding Schemes and Interactive Communication in Networks (Part 2 of 2)
10mRevisiting Network Coding for Warm Blob Storage
01h 14mDistributed and concurrent optimization for machine learning