FM Stereo "noise" and its removal

Just very recently the FM stereo tuner and its aerial that I use to capture BBC FM brodacasts off-air has developed some noise in its signal. I am presuming that this is something to do with the kit resolving (or otherwise) the stereo signal because if I switch the tuner to mono the noise disappears (but of course I am stuck then with a mono signal). Here is an image of the spectral analysis of the noise (taken from a quiet, announcerless, period prior to the show), showing a sharp peak at 19 kHz. The same noise is present on R2, R3 and R4 BBC FM.
Noise spectrum analysis.PNG
What is the best way to deal with this noise, in Audacity that is - I realize that the real best way is to fix the problem at the source hardware?

I have retained the noise sample in a project and hope to get a good “clean” bit of the noise from R3 later in the week, for future noise removal.

What I have been doing, following on from a tip I seem to remember a while back, is amplifying the noise signal for the sample prior to noise removal on the unamplified signal. Using the default NR settings the unamplified NR still left some audible chirping in the signal. A 10 dB amplification removed the chirping, but also left the signal a little flat. With some experimentation I settled on 7 dB as a good compromise. I do realize tha amplifying the “noise” also amplifies the floor, so I will lose a bit of wanted signal across the spectrum.

So my question is: is what I am doing good practice, or should I rather be experimenting with the settings in NR? If so are there any recommendations for NR settings that would work well with the noise profile I have?


For spikes like that you’ll be wantin a “notch filter”…

Audacity notch filter demo.gif

You can get a copy David Sky’s Notch filter plug-in here …

BTW if you increase the “size” (currently 512) on the frequency analysis, the spikes will get narrower and you’ll get a more accurate value for the frequency of each spike to put into the notch filter.

BTW2 try “log” instead of “linear” axis on the frequency analysis and you’ll be better able to see the spike on the far left, which is probably audibly the most conspicuous.
this spike (red) will be more visible if you select 'log' axis.jpg
BTW3 19KHz is inaudible, (unless you’re a dog) …

In FM stereo broadcasting, a pilot tone of 19 kHz indicates that there is stereophonic information at 38 kHz (19×2, the second harmonic of the pilot).
The receiver doubles the frequency of the pilot tone and uses it as a phase reference to demodulate the stereo information.

You have some other problem. 19.000KHz is the frequency of the broadcast stereo pilot tone needed to decode the stereo signals. It’s injected into the show at 9%-10% of the total loudness (reducing the show volume by 10%) and is always there in a stereo show. That’s what lights the little [STEREO] light on the front of the tuner.

I would be shocked if you could hear it. In most cases there is a filter for it in the receiver and it never makes it to actual sound – although that’s optional.

It’s extraordinarily difficult to make a noisy FM show that didn’t used to be noisy. The whole system is set up to make noise go away. Remember that picture of Armstrong’s experiment where he had a visibly sparking machine running next to an FM receiver and the receiver didn’t even know it was there.

There are some interesting ways of messing up, however. I have an older FM receiver which started giving noisy shows and it turned out to be internal electronics failure. I flipped the receiver over to AM and instead of the local AM station, I got almost 100% audio garbage. Exit one tuner.

External antennas do not live forever. Aluminum antennas develop a coating of aluminum oxide which is a dandy insulator. Since radio signals travel over the surface of the antenna, that’s the signal in the trash bin. If you make FM signals weak enough, they do get noisy as the detectors fail to have enough signal to reconstruct the show. This is the car radio in the third street tunnel just before the signal vanishes completely.

And we shouldn’t completely ignore the possibility that the station is broken. That’s not likely, but has happened. It is intended that the on-air operators listen to the air show, but if you’re listening to the mixing desk by accident, you can be off the air completely and not know.


Thanks for your inputs Koz and Trebor,

You’ve made me remember that my aged Nakamichi cassette-deck also has a button to filter the FM pilot tone. And yes my ageing ears certainly won’t hear the tone - especially not over the encroaching tinnitus :slight_smile: . I’m guessing that a strong signal at 19 kHz whilst inaudable could still trip the Audacity meters making them effectively over-read, right?

Actually I’m only using internal antennas since I moved house a couple of years ago. We live close enough to the re-transmitter to get a good strong signal. But that was a good point

The station doesn’t appear to be broken - I have another tuner in the same room. It is one of those little compact boombox "hi-fi"s and that’s still resolving FM stereo with no fuzz. I guess I could resort to using that, but the only outputs are to speakers and headphones - there’s no direct output via Aux RCA sockets unfortunately.

It will be somewhat academic in a couple of years of the UK government gets its way and switches off the currently wonderful FM BBC wavebands in favour of the “supposedly superior” digital DAB broadcasts. At which time it will definitely be new tuner time - or just record from the BBC website.

@Trebor: I have another recording to make off R2 this evening. I will experiment with the notch filter on that over the weekend.


The 19KHz signal peak is @ -42 dB. If you were using a gate with a threshold of -50dB the -42 dB 19KHz signal could prevent the gate from being activated even though the 19KHz tone is too high pitched to be heard.

You could remove the 19KHz peak with the notch filter, or low pass equalization.

I’d be more concerned with the peak on the extreme left (below 1000Hz). it is -70db which is very quiet but still may be loud enough to be heard.

If you post a bit of the “silence” we can hear for ourselves what the noise it is like.

If the noise is present on all channels but only on one radio receiver then it may be the audio amplifier of that receiver which is adding the noise, (not poor reception).

The 19KHz signal peak is @ -42 dB.

Which means there is some filtering going on, or you just have the show turned down. The pilot tone is maintained at -20dB compared to the peak FM signal (10%). And yes, what it would look like in Audacity is “thick” waves.

And I can’t seem to generate an example for illustrations.


ok - here is my 1 second noise sample, exported in 32-bit float WAV

Yes, as Koz pointed out earlier this may well be down to a component failure in the tuner. It is 18 years old. And the fault occurred after I turned the power off on the unit while I was away over the weekend - so I guess powering up again could have triggered a component failure.


interval between bursts of noise is 0,02sec ie 50Hz.png
The noise is bursts are 0.02 seconds apart, i.e. 50 Hz , the mains frequency (in the UK).

A possible explanation is that the components (capacitors) which smooth the power supply
in your 18 year old audio device have started to fail and 50Hz ripple is now making its way into the circuit.

Ripple is undesirable in many electronic applications for a variety of reasons:

  • The ripple frequency and its harmonics are within the audio band and will therefore be audible on equipment such as radio receivers

It could still be interference though, from a mains powered appliance in your home , (maybe something with a motor: central heating pump/valve ?, vacuum cleaner?, fridge ?).

Noise Profile.wav.

I know what that is. That’s my neighbor’s million-year old refrigerator. Back when my off-air tuner was dying, I could tell precisely when his chiller pump went on. I actually walked around the neighborhood with a portable radio and did signal strength analysis on a map. It centered on his kitchen.

Unfortunately, he’s between me and all the stations on Mt Wilson.

The tuner failed and the local station started posting high quality bitstreams on-line so I stopped worrying about it.


I think you could well be right Koz. :frowning:

Because after an afternoon spent with two other good FM radios moved around the house - still getting the same noise on both those radios in various rooms - and with either of them plugged to my recording set-up => “noise” still there.

But blow me, having now gone back to the original FM tuner plumbed in to my Edirol=>PC the “noise” has gone away, as I sit monitoring Radio 3 (the BBC classical staion where noise is most noiceable). It’s gone as mysteriously as it appeared a few days ago. I know it’s not my central heating pump as it’s coming on and off sufficiently during the day, in fact kicked on just now and no noise came back. At this time of year am I daft to suspect some neighbour’s fancy Christmas lights?

Anyway I’m keeping my fingers crossed for now - I do still have my Thursday recording from R2 that still has the “noise” so over the weekend I will try David Sky’s notch filter that Trebor suggested.

Thanks for all your inputs on this, I’ve learnt some good stuff from this.


If the noise is intermittent on one receiver then we can exclude ripple, as that is never spontaneously cured.

Sorry but a notch filter won’t remove that type of noise: a notch filter will remove pure monotones like the (inaudible) 19KHz which are a narrow peak on the spectrum …
FM 'noise profile' spectrum (size 16384), 19KHz is narrow peak (monotone)-.png
However your FM “noise profile” consists of bursts of “shush” noise which occupy a broadband* of frequencies: it’s essentially brief pulses of white noise every 0.02 sec. Broadband* noise is impossible to remove because it occupies almost all frequencies: frequency-wise there is no way differentiate that type of noise from the signal, (the music).

[* nothing-to-do-with-internet “broadband”]

Late in the day, but a note about the effects of surface oxidation:
Uniform surface oxidation of aluminium would have no effect on surface resistance - it just moves the surface down by a very small amount. Even the worst sort of pitting you might get has little effect on a 100-Mhz FM external antenna will have minimal effect on the performance, because the thickness of the bars is chosen for stiffness and RF impedance reasons rather than for the surface resistance. Typically the initial end-to-end resistance is less than 0.5-Ohms initially, and a factor of ten would make minimal difference to performance.
What can make a difference is corrosion of contacts - which is repairable.