Hacker News Reader

Comment by rb2k_ | Hacker News Reader

> Digital can be maddening because it alternates between crystal-clear and OFF, which ultimately I think is worse. Some streaming services will also decide to auto-next-song after the slightest glitch which is annoying too.

I wonder if there are audio codecs that work similarly to wavelet based video codecs (DWT instead of DCT) where you could have stepwise degradation.

View on HN

There is somewhat similar concept of Bitrate peeling [1] in Ogg Vorbis. You can drop portion of the stream without transcoding and still be able to decode original audio at lower bitrate. With proper over-the-air modulation and multiplexing scheme (like ADSL interleaving), one should be able to have graceful degradation of digital audio.

[1] https://en.wikipedia.org/wiki/Bitrate_peeling

Bitrate peeling is a beautiful concept, it's too bad that it was never implemented.

Problem is the fading. You could drop the bitrate to boost the link margin a few dB, which works fine for SATCOM, but it won't help with deep fades. Putting two antennas on the car for diversity reception would work much better; surprised no one is doing it.

Only cheap car radios do not do diversity. A lot of cars have three antennas and three or more tuners. You do not see all antennas, with a lot of cars you do not see any radio antenna.

Ha ha, I guess my cars are too cheap. I have never had a car with more than one VHF antenna.

The original DAB codec had more error protection for the more important parts of the audio data (unequal error protection), so while not the same it would give a more gradual reduction in quality instead of a sharp drop-off. DAB+ doesn't do this though.

It's a broadcast, you don't know what is the bandwidth of receiving device so you can't adjust accordingly.

The broadcast equivalent would be to transmit more error correction bits per capita with the important bits (i.e. those present in the lower constant bitrate stream). This is implemented in codec2 or the related modem system, I think.

Nah you can do something like:

1. Sort your bits by importance. 2. Encode them as a real number using arithmetic coding. 3. Transmit that real number in an analogue way.

That means the more noise you have in your signal the fewer bits you receive correctly.

There are probably much more clever ways to do it than that but I don't think much work has really been done on it unfortunately because it's quite complex.

Reply on news.ycombinator.com