CCSDS-LDPC-coding-scheme.md

@@ -47,6 +47,23 @@ TODO : add plots for codes with k=16384 whenever the performance issue on the GM
 
 Note that these simulations were executed with an iteration count of 200 and BPSK modulation, the same parameters as the reference simulation in the 130.1-G-2 document.
 
+## Execution time performance
+
+These numbers were achieved with a decode *iteration count of 10*. You can assume that throughput is directly proportional to the number of iterations performed. Thus, you can easily extrapolate the throughput depending on iteration count. These benchmarks were performed on an *i7-3770 @3.4GHz CPU* with *single-threading only*.
+
+| LDPC Code | codeword decode time (us) | Throughput (Mb/s) |
+| ------ | :------: | :------: |
+| AR4JA K=1024, R=1/2 | 370 | 2.77 |
+| AR4JA K=1024, R=2/3 | 255 | 4.01 |
+| AR4JA K=1024, R=4/5 | 196 | 5.23 |
+| AR4JA K=4096, R=1/2 | 1536 | 2.67 |
+| AR4JA K=4096, R=2/3 | 1081 | 3.79 |
+| AR4JA K=4096, R=4/5 | 831 | 4.93 |
+| AR4JA K=16384, R=1/2 | 6208 | 2.64 |
+| AR4JA K=16384, R=2/3 | 4282 | 3.83 |
+| AR4JA K=16284, R=4/5 | 3245 | 5.05 |
+| K=7156, R=7/8 | 1251 | 5.72 |
+
 ## Known limitations
 
 Construction of the generator matrices needed for LDPC encoding involve matrix inversions and matrix product. On codes with information block length k=16384, matrix inversions and matrix products take a lot of time (10s of minutes) due to the large size of the matrices involved. This remains to be optimized.