-
Notifications
You must be signed in to change notification settings - Fork 38
Benchmark results
Oona Räisänen edited this page Jun 10, 2018
·
16 revisions
Redsea's performance is platform-dependent, not only its speed but also the percentage of groups it will be able to decode from noisy signals. This could be due to small differences in the liquid-dsp floating-point algorithms for ARM vs. x86.
These results were obtained by running a batch of test MPX recordings through redsea v0.17.0 on two different platforms, a Raspberry Pi 1 (Linux, 700 MHz ARMv6) and a 2013 MacBook Pro (macOS, Intel Core i7). The results refer to the percentage of groups decoded, and the percentage of used CPU time versus test file duration.
Pilot SNR is used as a measure of noisiness because it's relatively simple to calculate. Redsea does not use the pilot tone.
| Test file | Pilot SNR | % groups ARM | % groups i7 | %CPU ARM | %CPU i7 |
|---|---|---|---|---|---|
| vikerraadio-171k.s16 | 51 dB | 99.8 | 99.8 | 44.3 | 0.6 |
| raw-yleyksi-171k.s16 | 36 dB | 99.1 | 99.5 | 37.4 | 0.6 |
| puhe.s16 | 30 dB | 2.7 | 63.1 | 42.8 | 0.6 |
| raw-ylesuomi-171k.s16 | 26 dB | 64.5 | 94.0 | 43.2 | 0.6 |
| nova.s16 | 21 dB | 0.0 | 8.5 | 49.7 | 0.6 |