Comment by philpem | Hacker News Reader

Comment by philpem | Hacker News Reader

Parent Full thread

The problem is twofold: "graceful degradation" and the effect of digital compression.

Digital schemes like TETRA and DMR compress the audio and add error correction. The audio bitrate for DMR is somewhere around 2400 Baud. This means a lot has to be thrown away. As a result, some things confuse the codec (usually AMBE). I have Scottish and Irish friends with strong accents; AMBE makes their voices both indistinguishable and unintelligible.

The second problem is what's sometimes called "the cliff-edge effect" or "graceful degradation". An FM signal degrades progressively as the signal gets weaker. If there's a temporary 'fading', you'll get a bit more hiss for a second or two. An experienced dispatcher or radio operator will often be able to "hear through" the hiss.

Digital signals don't degrade gracefully. The audio quality is perfect up to a point, but after that point things fail quickly. You get corrupted, missing or glitchy audio, loss of control packets (more lost audio), loss-of-sync (no audio until the next SYNC burst, ~1-2 secs)...

Incidentally, analog speech scramblers have similar issues. Rolling Spectrum Inversion scramblers send an FSK data-burst every time the seed (ephemeral key) changes. If your radio misses one of these bursts, the rx/tx audio will be screwed up for a second or two.

I still haven't fathomed why ETSI chose AMBE+2 for DMR. It's a codec made by an American company for compressing American English speech and it works reasonably well for that purpose. However, most speakers in my native language sound as if they were drunk or having a stroke.

Even Motorola has recognized that problem and added options in their programming software that makes the DSP boost higher frequencies before feeding the signal to the vocoder.

As to the graceful vs cliff-edge degradation of signals I've yet to see real comparisons where both analog and digital radios use the same frequency, radiated power and bandwidth (with co-channel interference!). With digital systems, the degradation will surely be much more noticeable, but the real question is whether the critical signal level is above or below that of the analog system.

> I still haven't fathomed why ETSI chose AMBE+2 for DMR. It's a codec made by an American company for compressing American English speech and it works reasonably well for that purpose. However, most speakers in my native language sound as if they were drunk or having a stroke.

A lack of ethnic diversity in vendors and standardization committees, which has also led to issues such as the "racist soap dispenser" (https://gizmodo.com/why-cant-this-soap-dispenser-identify-da...).

When most or all testing is exclusively done on White US-english male persons, issues such as a lack of sensitivity/quality for anything diverging from that norm only crop up when it is way too late to fix.

It's more than purely testing[0], it's a lack of self-awareness. I remember seeing marketing / engagement material targeting worldwide customers and using American football as the illustration. Wrong kind of football if you're aiming beyond NA.

[0] Not to detract from the skin colour issues we've seen countless examples of, like the example you mentioned or the Google Photo fail https://news.sky.com/story/google-photo-app-labels-black-cou...

I expect it's partly due to the market ecosystem when DMR was created. The standard is codec-agnostic, but in practice DMRA (and thus all the manufacturers) settled on AMBE. If APCO P25 predated DMR, that'd at least partly explain the codec choice.

I'd absolutely love to see a Codec2 based DMR radio, even if it were a firmware mod to an existing radio.

As for the degradation - same. The only data I've seen is from radio manufacturers and the DMRA. Practically it's going to depend on the specific radio, the antenna and so on. You could measure sensitivity as dBm vs error rate for digital or dBm vs SINAD for analog and plot the two together.

Once you cross the line of what the ECC/FEC can handle, things tend to break down very quickly. Most DMR radios will cut out entirely if they see a lot of Post-FEC uncorrectable bit errors.

Reply on news.ycombinator.com