Decoding MPX without pilot tone

[Otuxam3]

Hello there !

I saw the last update and actually thought about something to share here. After moving to a new location last year, I've started listening to local radios and I was surprised that one of them was broadcasting in stereo without a pilot tone! As much as it's weird and against probably all known specs in the FM stereo modulation world, I find it could be an interesting challenge to try and demux it anyway so here's my thought about the process...

Normally, to properly demodulate stereo and keep it in phase between M and S subcarriers, an initial calibration on the transmitter must be done, that's a thing. On the receiver side, being at 19.0000000000000kHz and 38.0000000000000kHz is nearly impossible due to several factors, having the center frequency of the stereo subcarrier being exactly the double of the frequency of pilot tone is of great help though, so a dozen Hz offset isn't a problem...

Now here is the trick, the pilot tone is totally absent from my recording, not even attenuated. How to recover stereo this way? A first idea was to just pick 38kHz but phase would be constantly rotating, resulting in weird effects. Second possibility : find the center frequency. Very few stations I had the chance to visualize the composite signal used lower side band for the stereo subcarrier and when they did, it was with a pilot tone. But in this current case, I got both side bands. There are methods described on internet about how to retrieve the center frequency of a DSB-SC signal. The last problem would still be the phase which would result in bad stereo separation, however it would be locked and a manual setting could allow for fine tuning.

I still hold the recording if you want to tinker with it, my local law allows for recording and sharing of public broadcasts.

[windytan]

Wow, what a peculiar find! I'd be interested in seeing the signal if you have some clip with stereo content.

I wonder, has anyone working for the station ever been test-listening to the transmission and noticed that the stereo is not working? :)

We could use the squaring method to estimate a carrier for the DSB-SC, but it's quite computationally heavy because I haven't been able to get bandpass filters properly working with liquid ( #3 ). Convolving with a shifted kernel always produces some distortions for me. So I just do down shift, LPF, up shift.

Normal path (current):

1. Shift -19k
2. LPF
3. Shift +19k
4. PLL
5. 2x frequency
6. Use as carrier to demodulate stereo

Pilot-free path (draft idea):

1. Shift -38k
2. LPF
3. Shift +38k
4. Square
5. Shift -fmod(76k, Nyquist)
6. LPF
7. Shift +19 kHz
8. PLL
9. 2x frequency
10. Use as carrier to demodulate stereo

[Otuxam3]

Sorry for the late, here is the file (only available 30 days).

I couldn't tell from the recording if there's a fully panned source though...

[windytan]

Merci beaucoup, I can hear a subtle stereo effect in the background music. However it sounds like it's slowly wandering around using my solution. Need to think of another design probably.

[Otuxam3]

Have a great time learning french on ads and some outdated news haha.

I've been testing after sending you the file and from what I can hear, it's a constantly rotating phase on stereo subcarrier, which sounds like an old dying amplifier when the attenuation is the strongest. My only guess is that it doesn't seem to get the exact center frequency... yet. The hardest part is ahead now!

[Otuxam3]

I got an idea while randomly testing out stuff...

Okay so the best case would be to have a way to get the carrier back. Normally this would require the pilot tone which is the in-phase half of the stereo subcarrier. Here there's none but that doesn't mean we cannot retrieve it back from the subcarrier itself.

I've been testing on a few secs of AM radio as an example, adding a filter makes it behave just like the DSB-SC stereo subcarrier. For a bit of theory, what makes AM and DSB-SC different in the time domain after all? The fact you still can "see" the carrier, but the envelope that is the baseband takes the whole waveform height instead of only a half. So when it goes below zero, the carrier's phase is just flipped... but it's still there! That's how once a FFT is done the actual carrier frequency gets cancelled completely as there's a 50/50 of in-phase and reversed phase of the carrier in average.

To avoid this, we can (if filtering very well the stereo component) take only the absolute value of each sample, which will make a clear peak appear on spectrum at exactly twice the carrier frequency. With this, we can regenerate the carrier to properly demodulate content while locking phase to either 0° or 180°. From there... after reassembling the correct LR channels we'll have 50% chance of having the channels swapped.

To add some pictures to the words, here is the filtered subcarrier :

And here is the result of a simple absolute value calculation :

Doing such a process is indeed very destructive but if only used to guess the carrier frequency, then it's worth a try 🙂

In hope this helps...

[windytan]

Okay, I think I got what you mean. I compared the squaring method and the abs(x) method and got several dB better SNR using the absolute value. So it might be worth exploring! I haven't tried yet if the PLL would get a steadier phase lock for this one.