Introduction
Polar codes represent a significant milestone in information theory, being the first class of error-correcting codes proven to achieve the Shannon limit. Their adoption in 5G New Radio (NR) and Internet of Things (IoT) standards underscores their importance. However, the standard decoding algorithm for polar codes, Successive Cancellation (SC), faces a critical implementation hurdle: its inherently sequential nature leads to high latency and limited throughput in hardware realizations.
The SC Decoding Bottleneck
The SC decoding process can be visualized as a binary tree traversal. To determine the value of a specific leaf node (representing a bit), the decoder must recursively process Log-Likelihood Ratio (LLR) values from the root down to the leaves and back. This serial dependency means that traditional decoders often wait for neighboring results before proceeding, creating a bottleneck.
While combinational decoders—which use raw logic gates without internal storage—can reduce latency, they suffer from long critical paths. As the signal travels through numerous gates, the maximum achievable clock frequency drops, limiting overall performance. The challenge lies in pipelining this process to handle multiple codewords simultaneously without excessive hardware overhead.
Proposed Architecture: The CPPU
To address these challenges, this research introduces the Compound Pipeline Processing Unit (CPPU). This hybrid architecture strategically balances combinational speed with synchronous pipelining.
Key Methodological Innovations
- Sub-tree Merging: Instead of a purely sequential approach, the CPPU utilizes combinational logic for smaller sub-trees at the leaf levels. These are then integrated into a synchronous, pipelined framework for the higher levels of the decoding tree. This approach effectively shortens the critical path compared to standard merging functions.
- Simplified CPPU (S-CPPU): At the final decision stages, the authors replaced complex arithmetic units with a Simplified CPPU. By utilizing Sign-Magnitude (SM) representation, the design avoids heavy addition and subtraction, instead using simple logic gates and comparators to process LLRs. This optimization significantly reduces the number of Lookup Tables (LUTs) required on the FPGA.
- Auxiliary Registers: Strategically placed registers act as data buffers. They store intermediate LLR values, allowing the hardware to begin processing a new data packet while the previous one is still traversing the later stages of the decoding tree.
Implementation and Results
The design was implemented using a parametric VHDL approach, allowing for scalability across different code lengths ($N$). Testing was conducted on a Xilinx Virtex UltraScale (XCVU190) FPGA using an Additive White Gaussian Noise (AWGN) channel model and BPSK modulation.
Performance Metrics
- Throughput: The architecture achieved a peak throughput of 2672 Mbps for a code length of 16.
- Hardware Efficiency: The S-CPPU logic resulted in a 17% reduction in LUT usage, freeing up significant FPGA resources.
- Latency Improvement: For a 128-bit code length, the decoder demonstrated a 17.34% speedup over existing state-of-the-art combinational implementations.
- Quantization: The study identified 5-bit quantization (3 integer bits, 1 fractional bit) as the optimal balance between hardware footprint and error-correction accuracy (Bit Error Rate).
Conclusion and Future Work
The CPPU architecture successfully breaks the latency bottleneck of SC polar decoders by combining sub-tree combinational logic with pipelined processing. While the design offers massive gains in speed and resource efficiency, the authors note that routing complexity on the FPGA becomes a factor for very large code lengths ($N > 2048$). Future research will focus on hardware-in-the-loop testing to evaluate performance in real-world 5G base station and satellite communication environments.
Summary updated