Before anything else, thanks for the detailed explanation. I learned quite a bit from it and have a more solid basis for thinking what I had intuited before.
You're quite welcome.
Looking at the graphs reminds me of dropouts seen on 2" quad videotape. In essence, the algorithm is creating dropouts at the boundaries of frequency bands.
Good comparison. Yes, what's specifically taking place is that because of bit starvation, the IBOC AM codec couldn't even encode most of the upper portion of the real audio range (the range just beneath 3600 Hz). Consequently, with almost nothing in that range for the SBR logic to calculate any synthetic harmonics from, echos of that "gulch" appear multiple times within the synthesized SBR range. In every way imaginable, the audio is a complete wreck and a total loss.
I think that answered my question about error correction. That would make sense since an AM signal is so prone to interference. In theory, a digital signal ought to be able to ignore most of that but imagine what intense electrical interference could do.
Plus skywave at night. Plus adjacent channel interference.
Possible answers for why this is are not flattering to broadcasters: neglect, carelessness, lack of understanding, being cheap with bandwidth, etc. But more on that in a moment.
At least broadcasters can increase their internet streams' bitrates. IBOC AM comparatively is a bitrate straight jacket. It's capable of going as high as 60 kbit/s from what I understand, but the sacrifice in FEC is said to make it too tetchy.
KOA sounds a little better than KHOW as far as digital compression goes but it still sounds like it's about a 32-40 kbps bitrate that's in use. maybe even less. What the {bleep}? Even worse, I hear these same effects on KBCO, which is an FM station - and I mean its analog signal. It's appalling. Is this some effort to get the sound on the stream to match the sound on the radio?
Appalling is the right word for that sound, but what you're hearing isn't actually lossy compression. I understand why you thought so, however, since very low bitrate lossy codecs produce a similar tonality to this stuff. What you're hearing, on both of those stations, is actually the accursed sound of Voltair. You can read about it this thread I started two years ago when I too wasn't yet clear on what was causing it:
https://radiodiscussions.com/thread...e-watermarking-is-this-kfi-and-others.756462/
Basically, Nielsen's PPM ratings watermarking encoders add station ID information in the form of modem sounds to 10 very thin slices of the audio spectrum between ~1 and ~3 kHz. Imagine something like FSK tones, with one at 1 kHz, the next at 1.2 kHz, the third at 1.4 kHz, going all the way to the 10th at around 3 kHz. Well, you don't want listeners to ever hear those. So Nielsen's PPM encoders only add them at very specific moments when they sense that other, existing sounds at those frequencies would psychoacoustically mask them. The problem with this approach is that, during any periods of time in a station's broadcast where there are no exiting sounds of sufficient loudness at those 10 frequencies for psychoacoustic masking to take place, the Nielsen PPM encoders have to stop sending their streams of 1's and 0's and wait for another moment in time to arrive where sufficiently loud sounds
do exist at those frequencies to enable psychoacoustic masking once again. As soon as that happens, the encoders resume sending their 1's and 0's from where they left off. This process repeats indefinitely, but depending on how spectrally dense vs. open the station's audio is on average (imagine the spectral density of heavy metal versus speech), the opportunities for masking can be so few and far between that it can take minutes before even one iteration of a station's ID can be fully transmitted to listeners' PPM meters. Meaning that if any listener listens for less time than that, his meter doesn't log the station as having been listened to. Other listeners meanwhile might listen for as long as it takes for the full station ID to be conveyed a few times, but thwart their meters' ability to fully hear any of those iterations with intermittent background noises. So someone at a Telos Alliance sub-brand company called 25-Seven came up with the idea of creating a processor (the Voltair) that would apply noise-gated upwards compression to the 10 aforementioned ultra-thin frequency slices to make Nielsen's PPM encoders almost always see audio in those regions that's sufficiently loud for successful psychoacoustic masking. The result, as shown by this
high-resolution close up of the Voltair's LCD display, is that its upwards compression amplifies so many psychoacoustically non-masking soft sounds so often that a PPM encoder can be tricked into near-continuous insertion, going from taking several minutes to send a station's complete ID message to just seconds.
But there's one small problem with that. Just as the ear perceives a continuous sinewave at one fixed frequency as "a tone," it perceives a dramatically amplified ultra-thin slice of spectrum during speech or music (i.e. 0.95 kHz to 1.05 kHz being 12 dB louder than everything on either side) as something akin to a variable-amplitude tone, or in audio processing parlance, filter ringing -- the sort caused when a phase-linear bandpass, bandstop, lowpass, or highpass filter has a vertical skirt/slope. And this, getting back to your "apt misidentification" of Voltair as low bitrate lossy audio, is why the Voltair sounds so similar to a bit-starved lossy codec. Lossy codecs also use razor thin, vertically-skirted EQ bands; only in their cases, to de-amplify (down to -infinity dB) all audio at specific frequencies for specific lengths of time (i.e. the sounds the codec believes would be inaudible anyway due to psychoacoustic masking, and that can therefore be silenced to save bits). Lossy codec artifacting sounds like "tinkling ringing" because it consists of thousands of these micro-thin slices of the audio spectrum being notched out by vertically-skirted bandstops that turn on and off at light speed. Voltair, by comparison, sounds like "reverberant ringing under a damper" because it consists of 10 equally thin vertically-skirted slices of the spectrum having their audio dramatically upward compressed (loudened/thickened) with a gate to keep the compression from pulling up the noise floor itself.
Here are spectral images of your KOA and KHOW airchecks:
https://files.catbox.moe/0q4on4.jpg
https://files.catbox.moe/toch0f.jpg
Note the 10 "Freddie Kruger claw marks," as I phrased it in my 2022 thread, throughout each one. They're identical to those seen in the KFI spectral image in that thread's first post (
direct link).
For a real eye-popping visual of this stuff, take a look at this audio clip I recently grabbed from KLAC 570 AM while they were airing baseball. Baseball game crowd noise, when the announcers aren't talking, has a very soft, gaussian spectral distribution that allows Thor's hammer tracks, as laid down by their Voltair, to stand out like sore thumbs. I put CoolEdit's spectral view into "HD mode" for this one. Just ... yuck!
https://files.catbox.moe/yxdzx5.jpg
https://files.catbox.moe/cpsk4d.jpg (zoomed)
https://files.catbox.moe/29syen.flac (source audio)
(I widened the bandwidth 24 seconds in)
If you're interested comparing my market with yours, here is a three and a half minute audio tour of several FM stations in Los Angeles. I simply punched in their frequencies at random, off the top of my head. See which of them you can hear the Voltair in. The stations in this clip, in order, are KYSR, KPWR, KNX, KOST, KCBS, KLOS, KRTH, and KUSC, and I included deliniating beeps. Hint: only two of them are free of Voltair, to both my ears and CoolEdit's spectral viewer. (The amount of amplification a Voltair box performs is user-configurable, so each station has an audibly unique amount.)
https://files.catbox.moe/5junzv.flac
Quite possibly the worst instance of Voltair I've heard,
ever, is currently coming from KYLA 92.7 FM here in Los Angeles.
These people seem to have their Voltair turned up
past 11:
https://files.catbox.moe/vbyw1i.flac
In closing, this stuff leaves me speechless. I never believed I'd witness the broadcast industry so masochistically trashing its own audio, and so systematically. Almost the entire dial is unlistenable to me now, and the bitter irony is, it has become that way out of their desperation for listener metrics. Although there's a tonal similarity of this stuff to lossy codecs, I think that Voltair cranked up high actually sounds worse. The filter ringing produced by bit starved lossy codecs is spectrally diffused and impulse-like. But the sound Voltair creates is fixed at 10 frequencies and smears itself across the audio temporally. So in a sense, it sounds like having an entire
discordant chord worth of tinnitus. The sheer egregiousness of that, going by the KYLA example alone, or that unbelievable KLAC spectralgram, should be enough to start having licenses called into question. Don't you think?
