New data fuzzer

The main focus of my UTSC standard is reliability so I needed a way to test how it responds to bit errors. Randomly corrupting data is called "fuzzing" but I couldn't find a program that was easy to use so I wrote one in Liberty BASIC.

My program takes a file to be fuzzed, and another file containing high-quality random bytes. It uses 24-bit values from the random file to get byte positions to fuzz, and uses "mod 8" to get the bit position to flip. This means it can randomly flip bits in files up to 16 MiB.

You can fuzz anything you like, but I wanted to fuzz audio so I could see how it would sound when the signal is weak and bits are being corrupted. The sound worked on analog TV when the signal was too weak for the picture to come through, and I want UTSC to do the same. First I tried Opus files. They can handle some bit errors but they stop playing altogether if there are too many. And if the header is corrupted, they won't play at all. I added a header-skipping feature to my fuzzer but obviously a real-world signal could lose the header.

Then I got to thinking about WAV audio. It has a very small header (44 bytes) and the audio portion can withstand unlimited bit errors without stopping. Of course, you need the header to know the sample rate and format, but what if "best practices" were defined for UTSC that define a default WAV format? After some tests I found that 24 kHz mono 8-bit WAV files are a good compromise between quality and bandwidth. As I wrote on the LostCarrier.Online Discord channel today, "With 17% bit errors, it degrades like analog sound and fades into the noise rather than glitching." My code was off by a factor of 8, so I meant 2.125% bit errors.

I already specified Opus as the recommended audio format on UTSC, but it can't handle anywhere near enough bit errors to be reliable in bad conditions. Let me demonstrate with a 10-second clip from Syn Cole's "Feel Good" from NoCopyrightSounds.

This falls outside of YouTube use, so hopefully 10 seconds is short enough to fall under Fair Use, but if not I'm including the attribution and I'll gladly swap the clip for something else if the owner complains.

"Syn Cole - Feel Good [NCS Release]"
https://www.youtube.com/watch?v=q1ULJ92aldE
Syn Cole
https://soundcloud.com/syncole
https://www.facebook.com/SynCole
https://twitter.com/SynColeOfficial
https://www.instagram.com/SynCole/




Notice that with only 0.2% of the bits flipped, the Opus file is barely playable. In contrast, the WAV files still contain obvious music even with about 50% errors. I say "about" because since this is a random process, some bits may be flipped twice and be unchanged, so 50% is only a reasonable figure. We can assume it's very close to 50% because of the high quality of the randomness used.

These results make me think that 24 kHz mono 8-bit audio is the optimal format to use if you want to ensure audio reliability at low bandwidth. However, the bandwidth is much higher than Opus. With Opus at 48 kbit/sec, the audio takes about 5.7% of the channel bandwidth, counting overhead. Using WAV as I've described would take 192 kbit/sec, or about 19.24%. That's roughly 4 times as much bandwidth just to make sure the sound gets through.

UTSC offers 1 Mbit/sec of bandwidth. So with Opus audio, about 94% of the bandwidth is available for video compared to 80.75% when using WAV. It's up to the broadcaster to decide if losing 135.8 kbit/sec of video bandwidth is worth it. If extra-high quality is desired, it may not be.