Computer Architecture Project LDPC Belief Propagation Decoding with Multi-Threading and Hardware Acceleration Computer Architecture Project
Low-Density Parity-Check Codes Member of the error correcting codes family LDPC codes provide performance that approaches the Shannon Limit Block and Convolutional Codes. Belief Propagation Algorithm for decoding The technique that wins in all three objectives is the goal of the computer architect
Belief Propagation Algorithm LLR0 LLR1 LLRn-1 Variable Nodes Check Nodes Log Likelihood Ratio (LLR): A measure of how confident the decoder is of the received bit being one or zero. One such technique is….. Research has predicted that Branch Prediction will have increasing importance as pipelines deepen and issue widths increase The LLR messages will propagate between the nodes during each decoder iteration. The LLR values of the Variable Nodes will converge towards the correct values.
Critical Computational Path Approximated as
One Algorithm Implementation
Experiment Locate the costly-loop and study the speedup on a multithreaded processor Experiment Hardware Accelerator Architecture
XInC Microprocessor multithreaded 16 bit RISC 8 threads realtime
Experiment Tools and Metrics Used supplied C-compiler to generate equivalent LDPC-decoder assembly code Instruction count as metric
Results Architecture Inst. Count Speedup 9611 Base case ST & HA 6958 Single Thread 9611 Base case ST & HA 6958 1.381 Multi-Thread 1384 6.944 MT & HA 1005 9.563