Can pocket FM radio output at 50kHz?

[Paul Uszak]

I've recorded the output hiss from a small battery powered radio operating inside a metal tin. From related posts, I can say that the radio is effectively shielded from receiving any FM transmission, and the resultant hiss is thermal noise in the early stages of the receiver.

I believe that the radio can output frequencies up to 50 kHz via it's earphone socket. Can this be correct?

My findings suggest that it can. When not constrained by handling FM audio, the input to the early receiver stages is thermal noise which (unconstrained) has an almost infinite bandwidth. The only constraint operating on it then comes from the radio's internal electronics. These will have spare capacity well above the audio range so as not to clip FM audio.

I've attached screen grabs from Audacity. This is the recorded hiss, and visually compared to a basic 10 kHz sine wave you can deduce how the hiss might reach 50 kHz, albeit at lower amplitudes.

I've also uploaded a spectrum analysis, and a spectrogram. These suggest Pink noise.

And finally, I've uploaded a zipped FLAC file of 5 seconds of the hiss should anyone have a real spectrum analyser. This would help greatly

Is there any reason to think that a pocket radio can't have electronics capable of handling frequencies up to 50 kHz..?

 

[Baluncore]

electronics capable of handling frequencies up to 50 kHz..?

Define handling.

The purple spectrum shows three spurs at 30kHz or below. The presence of those spurs above the noise floor demonstrates that it is not white noise.

The -3dB point for the spectrum is at about 5kHz. That is the bandwidth of the audio output stage.
At 50kHz, the noise is about 55dB below the audio. That is about 0.25% of the audio band and so will be swamped by the audio and the spurs.

 

[nsaspook]

Why would you want to use a radio when the circuit I referred to in the other thread can give you this?

Bottom: noise floor of the measurement system
Next: noise floor with amp on and noise diode off
Top: Noise at 10mA and 60mA of current in the noise diode respectively.

 

[Paul Uszak]

Why would you want to use a radio when the circuit I referred to in the other thread can give you this?

One word: practical expediency and risk mitigation (is that more than one word?).

1. Why build when I can use what I already have?
2. Most if not all of the design challenges for my radio in a tin apply equally to your circuit. It will need shielding (perhaps a sweet tin?). It needs power and output leads. The signal level looks low so perhaps additional amplification might be required, and hence it's design. What's a good physical circuit layout and grounding regime?
3. My analogue skills are minimal.
4. If you look at similar commercial hardware generators they're very complex. They require great analogue and EMI design and manufacturing skills. They also require sophisticated equipment to verify their correct operation. I know the radio works as I can see and hear it go "Shhhh".
5. The operational frequency of your circuit is inappropriate for my use case. (Thanks anyway.)

I intend to consume the radio output at a maximum rate of 96 kSa/s. In accordance with sampling theory, this should cover a bandwidth sufficient to go right up to my spectrum's shoulder at approximately 40 kHz (affording at least a 100kb/s random number extraction) . That's equivalent to the pink region I've highlighted in your spectrum. Not sure what's going on there - is the circuit even operating at that frequency? If it is, it's very quiet and perhaps indistinguishable from the other noise floors. The radio only ranges 25 dB around here.
6. The output spectrum's profile is unimportant to my use case. Any shade of noise will do.

I'm not categorically ruling out a dedicated noise circuit, but I suspect that most of them will still carry the above risks and limitations. So for the time being, I see the radio in a tin my least risky choice. A radio works pretty well for random.org after all.[/QUOTE]

 

[Paul Uszak]

Define handling.

I mean that the radio can produce an hissy output signal in the range 0 - 50kHz in the absence of any FM signal. That's what I'm taking from the spectrum, unless you tell me otherwise. I have little knowledge of sound engineering, and am relying on more experienced persons' feedback.

I don't understand your comment distinguishing audio from noise. The spectrum/graph is all noise. The radio's in my tin.

 

[nsaspook]

That's equivalent to the pink region I've highlighted in your spectrum. Not sure what's going on there - is the circuit even operating at that frequency? If it is, it's very quiet and perhaps indistinguishable from the other noise floors. The radio only ranges 25 dB around here.

That's not really a problem, you can sample the noise to produce your required output rate.

Some data on FM radio noise.
http://www.ciphersbyritter.com/NOISE/FM1ME904.HTM
http://www.ciphersbyritter.com/NEWS5/FMRNG.HTM

 

[Baluncore]

I don't understand your comment distinguishing audio from noise. The spectrum/graph is all noise.

To get a quick estimate of an RF filter profile I make a noise source by chaining three or four broadband MMIC amplifiers together with a 50 ohm input termination resistor to produce Johnson noise. The output goes to a spectrum analyser to check the noise profile is close to flat. Then I insert the filter between the noise source and the analyser. The change in the noise profile is the characteristics of the filter.

The spectrum of the output of an FM radio is the spectrum of the noise generated by the phase detector, multiplied by the passband of the detector and audio amplifier stages. You have no way of identifying what is noise profile and what is filter characteristic.

The shoulder of the audio output when used as a noise source is clearly the -3dB point. That shows the bandwidth is only 5 kHz, which is not really surprising for a cheap audio amplifier.

If you want a greater bandwidth than 5 kHz you must use a bandpass filter to extract and level part of the spectrum. The resulting noise signal will need to be amplified during the process. The maximum bandwidth is then the frequency difference between the -3dB points of the extracted band. I would select the band between 10kHz and 30kHz because flattening the spectrum there looks possible. The best bandwidth I could then expect would be 20kHz.

 

[Paul Uszak]

The shoulder of the audio output when used as a noise source is clearly the -3dB point. That shows the bandwidth is only 5 kHz, which is not really surprising for a cheap audio amplifier.

I don't recognise these numbers from my spectrum. Doesn't it show signal ranging zero to approximately 40 - 50 kHz? Let's not use the term audio as it's >> 20 kHz.

There's a real risk that I've miss-understood the signal analysis as I'm new to it. I uploaded a .FLAC clip of the signal with my initial post. If you have the resources to analyse the spectrum yourself, I'd greatly appreciate comments.

 

[meBigGuy]

40 dB down (~ -80 on the graph) means you have 100 times less voltage at that frequency than at the peak. dB = 20 log V1/V2

Normal frequency response corners are at -3db (half power). -33dB to -36dB is -3dB.

You have no idea what is causing the frequency slope, nor the spurs.

You are definitely building a toy. Period.

 

[Baluncore]

Let's not use the term audio as it's >> 20 kHz.

The spectrum you show is the audio output from an audio amplifier. Compared with the low frequency noise present, the -3dB point is at only 5kHz.
Even a 1 volt 100Hz audio signal has some 1GHz Johnson noise present due to in the resistance of the wires. It will probably be about 160dB down. That does not mean you can turn the 100Hz audio tone into 1Gbps random bitstream.

Why the -3dB point? See; https://en.wikipedia.org/wiki/Bandwidth_(signal_processing)

The problem with signals that are -3dB relative to the bigger signals is that they are hidden by those bigger signals.

Consider an AC coupled flat noise spectrum from 10Hz to 50kHz. If you use a comparator to sense the sign of the voltage you will get a rectangular noise waveform. You can sample that at a rate of about 25kHz and get a random bitstream.

Next consider a 10Hz sinewave oscillating between +1 and –1 volts. Now superimpose a 50kHz sinewave oscillating between +10mV and –10mV. If you feed that to the same comparator considered above you will get a 10Hz square wave output with noisy transitions every 50ms. That is not random. The 10Hz wave will take about 320us to pass through the 20mV nearest zero. Only during that very short period will about 16 cycles of the 50kHz signal be present in the random data stream. The rest will be a long line of consecutive 0s or 1s.

That is why the band of noise you convert to a random digital bitstream must be flat. If any part of the noise spectrum is more than about 3dB down on the rest of the band, it will not contribute to the random bitstream.

Until you understand the concept of the –3dB bandwidth you will be unable to make wise decisions about the bitrate of random noise.

 

[meBigGuy]

Not only do you need to understand the -3dB bandwidth conceptually, but you need to understand anti-aliasing filters and imaging (aliasing) versus sample rate and and the concept of decimation. But, you need to understand those even with a decent noise source.

 

[Baluncore]

@ Paul Uszak, your purple spectrum.png in post #1 is plotted with a linear frequency axis. You have the technology, please plot it again with a logarithmic frequency axis from 10Hz to 100kHz, it will then be easier to see the structure of the low frequency part of the output. While you are at it, move the zero dB reference level so that the highest amplitude part of the band is at zero dB. It will then be easier to interpret the spectrum and why it has the shape that it has.

 

[Paul Uszak]

This is the best I can do. I can either have a log scale and no 0 dB, or 0dB with no axes!

The FLAC file of the actual noise signal I used is attached in the first post if you have access to better resources...

Perhaps this helps a little, it's got a combination of both ...

 

[Paul Uszak]

...building a toy.

An AR15 and a Ferrari are toys, yet both can kill. You really shouldn't conflate "toy" with "it doesn't work."

 

[meBigGuy]

An AR15 and a Ferrari are toys, yet both can kill. You really shouldn't conflate "toy" with "it doesn't work."

LOL --- AR15 and Ferrari are precision tools, well engineered to accomplish specific functions. Engineered down to the last thread and every coating. How many engineers do you think it took to develop the paint process for a Ferrari? That people apply them in "toylike ways" is besides the fact. Trying to compare what you are doing to precision engineered machines is further indicator of how little you really understand about electronic signals. (And, you don't seem to want to listen with an open mind to some expert advice you have been given )

Compare a ferrari (well engineering random data generator) with a tonka toy (Not understood noise source of questionable randomness). Tonka toys work fine for what they are supposed to do.

What you are building as a noise source is a toy. Toys work fine for what's expected of them. I don't, however, think your expectations are in line with what you can actually accomplish with the tools you have limited yourself to.

Find some way to run auto-correlation on the flac file (rather on the data you try to extract from the flac file). Some toys are better than others. Maybe you'll get lucky.

http://www.ltrr.arizona.edu/~dmeko/notes_3.pdf

 

[Baluncore]

Attached is spectrum with a new vertical axis with zero RL.
Notice that the -3dB point is actually close to 2kHz.
That could give you about 1k bits per second.

 

[Paul Uszak]

To all Naysayers, Disbelievers and Merchants of Ennui ,

I offer two totally random images. One is my (Ferrari) image extracted from FM radio static. The other is from the binary expansion of e. Can you tell them apart?

Some toys can work pretty well...

(p.s. I'd have produced bigger images, but my prototype extractor only currently supports extracting 65k words. They both pass the simple ENT test, but of course that doesn't mean much with such small files. It's a promising start though.)

 

[Baluncore]

Please do not insert images into your post. Attach them so they can be downloaded if needed.
An image.png compression is not lossless, it merges adjacent pixels.
Your images are therefore randomised by lossy .png compression.
How many bits per pixel did you use to construct your image?

What data rate did you used to extract the data stream from the FM detector noise?

 

[meBigGuy]

lol I can create seemingly random pixel patterns with high auto correlation from simple formulas.

Yet another example of how little you understand what you are on about.

as for "Some toys can work pretty well...", I agree with that. I said as much. But they are still toys. Just depends on your expectations.

 

[meBigGuy]

Above: A plot of a series of 100 random numbers concealing a sine function. Below: The sine function revealed in a correlogram produced by autocorrelation.
(from the wikipedia autocorrelation article)

 

[nsaspook]

NIST Computer Security:
http://csrc.nist.gov/publications/drafts/800-90/draft-sp800-90b.pdf

Recommendation for the Entropy
Sources Used for Random Bit
Generation

 

[meBigGuy]

Just had a thought. I guess to the extent that the data sequence (not the flac file) is compressible, the entropy is poor.

If the flac file is smaller than the data (wav) file, then that proves poor randomness. That might not be true for a small sample though.
Just thinking that compression is a small (simple) measure of randomness. Maybe gzip -9 would be the best. I don't know too much about the different compression algorithms.

It's not a great test, since, for example, you cannot compress a jpeg much, and it obviously has issues. But as a first order test of compressibility it supplies a data point.

(actually, I expect that a jpeg compression of any image would be more random that what comes out of that radio).

 

[Paul Uszak]

Attached is spectrum with a new vertical axis with zero RL.
Notice that the -3dB point i...

How did you get the 0dB? I couldn't figure it out.

However, since you've annotated the graph with volts, shouldn't we be discussing the -6dB point rather than the 3? Surely with the sound card's input being relatively high impedance compared to the headphone socket's output, we should be looking at amplitude and not power?

 

[meBigGuy]

Maybe you should read a bit about dB, dBv, dBm, etc and what all that means. Try to determine why what you said was wrong (as much of it is).

 

[Paul Uszak]

Sorry about the images - I was trying to make a point ( and maybe performing a tune on a brass musical instrument owned by myself.)

PNG compression is lossless (honest), so the images are exactly as generated if you view them at 1 to 1 scale. They were constructed at 8 bits /pixel, indexed palette. Just bytes in a row. One of the images is the binary expansion of the digits of e, so whatever seemingly random pixel patterns with high auto correlation you can see, it's as random as current mathematics can establish. Sorry if some contributors don't like it, but there it is.

I sampled the radio signal at 96 kHz, 16 bits. This is sufficient to give me 96 dB fidelity and capture an output up to 44 KHz, which I think we have determined is present. Doing the maths, you get about 1.5 Mb/s of raw data. I extracted 65 k words to produce the other image. Using a simple compression test as a means of entropy mensuration, I get about 60% pure entropy from the radio. Not sure yet what I'd actually use in a production environment.

 

[Baluncore]

How did you get the 0dB? I couldn't figure it out.

The highest point on the graph is defined as 0dB. That point is at about 400Hz where there appears to be a marker.

You are new to this game. -3dB is the half power point. Power is power = energy flow rate. Voltage and current are impedance dependent.
-20dB is one tenth of the voltage and therefore also one tenth of the current in the impedance used, so -20dB is one hundredth of the power.

 

[Paul Uszak]

Following debate, I've tried some other software. I think that it's proved that my little radio produces output signal right up to 40 - 50 KHz.

If you look at fm.PNG, you'll see young person's music from an FM station. It drops off just after 10 KHz as expected. Then compare that with static.PNG. This shows output to about 40 KHz, dropping off rapidly through 50 KHz. This proves it I think. The radio stages after the FM receiver are capable of amplifying and propagating a signal up to 50 KHz. I would expect something like this as the circuit design must have tolerances and safety factors, and use readily available components that come in fixed ranges. Anecdotally, steel beams are designed in exactly the same way and usually can carry load some way past their design loading, up to the next available beam size.

P.S. For the worrywarts out there /here, have a peek at muted.PNG. This was with the radio connected, but with the volume dial only just turned on with no sound emanating. Just before you say that's noise on the sound card, floating.PNG is as inferred.

It's your white noise

 

[Baluncore]

Yes, 50kHz is there. It is at one 10,000'th the amplitude of the voice components near 400Hz.
It is your ability to understand the significance of the difference between -3dB and -80dB that is missing.

 

[meBigGuy]

It's unfortunate when we get a know-it-all that seems to think he has something to prove. The confrontational attitude and total dismissal of expert advice is pretty hard to sit by and watch on this forum. Generally the people with stuff to learn are willing to try and understand what is being said. This person sees it with nothing but ego.

This is not a debate. You are being told what is correct, and your refusing to accept it and not even trying to understand why it is true can hardly be called a debate.

I think the thread should be closed. It is going nowhere.

 

[Drakkith]

Thread locked for moderation.

EDIT: several posts have been removed and the thread will remain locked.