The National Grid can be used to convict you

reply to post by Arbitrageur

Are you comparing a BBC documentary to a science fiction show?

In this case, yes, I am afraid I am. It is disappointing to me that I have to do so, because I felt the BBC was a pretty informative and accurate network... until now.

Here's the deal in a nutshell:

If

• you make a phone call
  
• using a phone which has poor noise-cancelling ability
  
• from within a few feet of a power line transformer or older appliance that does put out a power line frequency hum
  
• in an otherwise perfectly silent area
  
• the power company is maintaining logs of that detail on their power fluctuations*
  
• the recording is of sufficient quality to detect the hum to that precise level

THEN AND ONLY THEN it may be THEORETICALLY possible for that hum to be recorded and matched up to the power company logs.

That's a lot of ifs. On the other hand, every telephone call made records at a minimum

• the time the call was placed
  
• the phone number the call originated from
  
• the phone number the call was placed to
  
• the duration of the call

all in easily-accessed data used for billing purposes, retained for extended periods in archives, and available via court order to any law enforcement agency with probable cause... the same restrictions as are on the power company's records.

* The power company probably does record frequency fluctuations, if for no other reason than to verify proper operation or potential issues. I cannot envision such detailed records being kept for more than a few days time, however, since after that they are of no further use to the power company. Perhaps a record of an anomalous reading may be kept longer, but that would be a serious hit-and-miss situation.

So, we are talking about how terrible it is that theoretically under perfect circumstances there is a slight chance that the power line signal might be used on some calls to determine the time placed... when all that information is available easily and more accurately from the telephone company.

Again, no.

TheRedneck

reply to post by TheRedneck


Well it looks like you are going to fill an encyclopedia after all. In any case, allow me to reply to a few things you said in your last post.

* The power company probably does record frequency fluctuations, if for no other reason than to verify proper operation or potential issues. I cannot envision such detailed records being kept for more than a few days time, however, since after that they are of no further use to the power company. Perhaps a record of an anomalous reading may be kept longer, but that would be a serious hit-and-miss situation.

Did you not watch the video or read the article I posted? Several forensics departments have a continuous log which go back half a dozen years or more. The reason is for intelligence purposes, such as tracking the time of audio and probably other things too. You really doubt that they would go to the lengths necessary to record all that data when it has recently been revealed that they record just about anything of relevance?

So, we are talking about how terrible it is that theoretically under perfect circumstances there is a slight chance that the power line signal might be used on some calls to determine the time placed... when all that information is available easily and more accurately from the telephone company.

I think you're missing the point, I doubt the main use of this technology is to track the time of ordinary phone calls, because you're right, they can easily look at the meta data if they need to. What they would use this for is for time-stamping video footage take on a hand-held video camera or other recording device which disallows them from easily pin-pointing the time of the recording. It's basically like a super high tech way of pulling meaningful information from a recording when all other techniques fail.

Originally posted by TheRedneck
reply to post by Arbitrageur

Are you comparing a BBC documentary to a science fiction show?

In this case, yes, I am afraid I am. It is disappointing to me that I have to do so, because I felt the BBC was a pretty informative and accurate network... until now.

Here's the deal in a nutshell:

If

• you make a phone call
  
• using a phone which has poor noise-cancelling ability
  
• from within a few feet of a power line transformer or older appliance that does put out a power line frequency hum
  
• in an otherwise perfectly silent area
  
• the power company is maintaining logs of that detail on their power fluctuations*
  
• the recording is of sufficient quality to detect the hum to that precise level

THEN AND ONLY THEN it may be THEORETICALLY possible for that hum to be recorded and matched up to the power company logs.

Did you watch the video?

They said the police record the frequency. So it makes no sense to talk about the power company maintaining logs (which I'm sure they do anyway, but the point is the police can use their own recording, without having to get any data from the power company).

Can't you hear music over the phone? The noise canceling feature doesn't cancel out the sound of music, does it? The 50Hz sound wouldn't be canceled out any more than a 50Hz note in music would be canceled out, though the frequency response of phones drops off at the lower frequencies like that, but it's still within their capability.

The area doesn't have to be otherwise perfectly silent. All you need is sufficient signal to noise ratio at the ~50Hz frequency.

So really the only valid point on your list is you need to be somewhere in the vicinity of a hum-emitting device, although I'd say "within a few feet" is probably not an accurate representation. It would really depend on the volume level of the hum, the sensitivity of the phone's microphone, etc.

reply to post by ChaoticOrder

I will admit I was somewhat flabbergasted when I watched a trace that looked like noise and was told it was a sine wave. Things like that tend to make me very skeptical and I did miss that point about the police maintaining the logs.

If the police are keeping a log of power fluctuations going back several years... well, I guess the police like to waste their time. You'll have a better chance of convincing me that this scheme is going to stop terrorism than you would convincing me that police in general are technologically savvy.

My apologies to the LEO's out there who do exhibit intelligence... I do know some of you exist. I speak of the profession generally.

What they would use this for is for time-stamping video footage take on a hand-held video camera or other recording device which disallows them from easily pin-pointing the time of the recording.

Perhaps, some day, they might get lucky and get a match... but that video was staged. It's not that easy. Even if they do get a match, would you trust your life to a procedure that has so many potential flaws?

I wouldn't.


reply to post by Arbitrageur

Can't you hear music over the phone? The noise canceling feature doesn't cancel out the sound of music, does it?

Music is not a continuous note. A 50 Hz line hum is a continuous signal.

Hold a continuous musical note for more than a few seconds and see if noise cancelling doesn't catch it.

The area doesn't have to be otherwise perfectly silent. All you need is sufficient signal to noise ratio at the ~50Hz frequency.

Precisely my point. You need a decent signal to noise ratio in order to be able to retrieve usable information. If the signal is low, then the noise must be much lower than that to give a good signal to noise ratio. The noise in an electronic communication is pretty much a set value based on the recording equipment. In addition, every source of AC hum is not emitting the exact same phase at the exact same time... it depends on the local resonances and reactances in those particular circuits, and more than one signal means the overall signal will contain all the signals. Which means that even the signal itself that you are looking for is distorted.

The signal to noise ratio may be excellent for voice where a line voltage ranges from +1 to -1 V, but for an inaudible hum at, say, +0.01 to -0.01 V it will be 100 times higher. Just because you can hear someone yelling across a crowded room doesn't mean you can hear someone whispering with their hand over their mouth across the same crowded room.

TheRedneck

reply to post by Arbitrageur


---

If you want to gather more information out of an analog signal used for forensic analysis purposes
one must INCREASE the sample rate (i say around 5 MHz for audio) and INCREASE the bit-depth
of individual samples (i.e. 32-bits per sample at 5MHZ) this prevents aliasing artifacts during quantization.

From there intelligence agencies, police, etc can analyze and obtain high resolution
mains-hum information for further forensic analysis and matching. So for you telephone users,
you're all SMEGGED...so don't use phones, use email (ENCRYPTED using an elliptic curve
algorithm rather than AES-256 bits) and trust only your closest mates!

---

And for absolutely utterly paranoid types, if there are KNOWN frequencies and or types of tonal
response in an audio recording that could be used to create a FREQUENCY RESPONSE chart
which could be used as a fingerprint to identify which microphone recorded the original recording.
This means that intelligence agencies tendency to have higher bandwidth recording gear could
obtain the frequency response curve of your telephone handset's microphone so that a positive
or negative match be made to one of their secret recordings of your conversation.
This DOES ASSUME that the "police" recording equipment has a MUCH HIGHER bandwidth
and far better frequency response curve so that the ORIGINAL handset's FR curve can be extracted
for pattern matching purposes.

This can ALSO be done with video recording gear!

The chromatic abberation curves of individual camera lenses,
the moire-pattern response of individual CCD or CMOS chips,
even the YCrCb to RGB conversion tables or CMOS chip Log tables
used to create the final on-screen imagery can ALSO be used to
identify individual video recording devices.

If there's electronics involved, they CAN be "watermarked" and tracked for forensic purposes!

Originally posted by AmberLeaf
Apparently the National Grid runs at 50hz, this signal may not be audible to the human ear in some cases....A slight buzz on a recording can give away the exact time the recording was taken.
Its pretty late here but i wanted to share this video which explains better what im going on about.....

The national grid doesn't run at a constant 50 Hz, it actually varies up and down depending on demands. You can make a timelapse video of street-lights and you'll see the flickering speed up and slow down. The national grid keeps track of this varying speed, so it then becomes possible to determine the time when a video was made.

Originally posted by stormcell
The national grid doesn't run at a constant 50 Hz, it actually varies up and down depending on demands. You can make a timelapse video of street-lights and you'll see the flickering speed up and slow down. The national grid keeps track of this varying speed, so it then becomes possible to determine the time when a video was made.

That doesn't exactly add up. First you talk about a time lapse video. Time lapse might be for example 1 frame per 10 seconds.

I think you mean high-speed video, maybe 1000 frames per second instead of the normal 25-30fps. Then you could see the variation.

But who makes videos at 1000 frames per second with a high speed camera? Almost nobody.

Looking at a normal 30 frames per second video trying to see variations of a few thousandths of a hertz variation in 50 cycles per second is not possible.

However using the sound track of the video might work, if the buzz is audible, as it's not split into 30 samples per second like the visual part of the video.

Originally posted by AmberLeaf
Apparently the National Grid runs at 50hz, this signal may not be audible to the human ear in some cases....A slight buzz on a recording can give away the exact time the recording was taken.
Its pretty late here but i wanted to share this video which explains better what im going on about.....

Recordings from my home studio don't have any buzz or hiss.

Haven't watched the vid yet but will. Very curious what they would key into in the hum.

reply to post by TheRedneck


Let me first state that it is obvious that in order for this to work the microphone needs to pick up the hum, and the recording needs to be accurate enough so that changes in frequency of the hum are easily picked up. And that most of the time, particularly with cell phone calls, it would be easier to simply get phone call data from a phone company. So it won't work with just any phone call. However it does seem completely plausible that if there is an audio track, not necessarily from an actual phone call, where there is mains hum present, then this technique would work. It has actually been used to verify a timestamp in a court case as you can see on the previous page.

But the initial conversion to digital format is relevant. In order to digitize a signal, that signal must be broken down into quantized packets, in the process losing some of the original analog data. For voice transmission, or even more difficult signals, sufficient bits can be employed to minimize the information loss, but when the signal component is inaudible that loss can easily be enough to remove any usable information.

What you are effectively saying is that the 50 Hz signal must be loud enough. I agree.

The phase shift will affect the resonance of the signal, however, and would cause interference with any power line component introduced, which would cause a slight frequency shift in the reading due to the malformation of the interfering waveforms.

This doesn't make any sense. You are using terminology you do not understand.

No matter the combination of resistive, inductive, or capactive elements, if the input is a certain frequency sinusoidal then all voltages, currents, and sum of these elements will also be sinusoidal of that same frequency. Since we are measuring frequency, the fact that different elements will have differing phase angles or may be in resonance is completely irrelevant.

The bandwidth affects the accuracy of the signal. Just as there is more detail in a 10 MPx camera image than in a 1 MPz image, there is more accuracy and thus usable information in a higher bandwidth signal than in a lower bandwidth signal.

What matters is how much the frequency of the ~50 Hz signal can be expected to vary and the corresponding accuracy required to measure changes in it. The 50 Hz signal is not sending any information, it is just a sinusoidal, therefore it is inaccurate to state a low 'bandwidth' (I'm not even sure this is the right word to use in this context) means less information - it isn't carrying any in the first place.

Noise can be audible to humans. Noise filtering involves detection of constant signals (such as a 50-Hz hum) and removing that particular frequency, as well as using feedback recognition techniques to remove noise caused by feedback.

A 50 Hz hum isn't constant though. It's only there if equipment generating a hum is present. It would only be cancelled if the noise cancelling technology is designed to filter ambient sound and is adaptive.

I thought I would chime in as I know what I am talking about lol.

Firstly, the 'noise' plot that redneck is talking about is actually Hz/Time rather than a plot of the mains itself. It is true that if the mains looked like this there would be trouble.

Secondly, the majority of 50Hz pickup on recordings is recorded via electrical noise not audible noise. This would still work in a silent room. In most cases phase offsets won't cause any problem for this sort of analysis where the zero cross points per interval are counted on the bandpassed signal within a within a time window. You then slide the window up on the time axis and repeat to get the frequency variation.

Thirdly, mains noise is everywhere and present on most audio recordings to some extent.

Forthly, if the call was recorded via analog means there would be a second 50Hz signal present that might make analysis more difficult.

Fifthly, UK mains varies by more that what is mentioned. +/- 0.2Hz is quite common. You can view a trend online National Grid

reply to post by C0bzz

This doesn't make any sense. You are using terminology you do not understand.

No matter the combination of resistive, inductive, or capactive elements, if the input is a certain frequency sinusoidal then all voltages, currents, and sum of these elements will also be sinusoidal of that same frequency. Since we are measuring frequency, the fact that different elements will have differing phase angles or may be in resonance is completely irrelevant.

Let me see if I can clarify my statements. I obviously was not clear.

The difference in time between identical wave points on a 49.99 Hz and a 50.01 Hz (.02 Hz bandwidth) is approximately 0.000008 seconds. In order to detect this accurately would require a sampling rate of 125kHz. Typical sampling rates are 44.1 kHz, about one-third of the minimum rate required to make a differential measurement based on wavelength and about one-sixth the minimum rate to give reliable results..Thus the audio track cannot be used to provide a peak-to peak, trough-to-trough, or zero-crossing analysis.

The only other way to measure frequency, and the one commonly used, is to compare voltage differentials with respect to time at the zero-crossing points of the wave. This works with sinusoidal wave patterns fairly well, since the maximum differential voltage rates can be predicted from the frequency. That becomes impossible when phase shift is introduced, because the waveform is distorted from a true sine wave and the voltage differentials become impossible to predict without exact knowledge of the waveform itself.

In an attempt to simplify, it is completely possible with very little stray inductance/capacitance for the leading edge of a 50 Hz signal to give a voltage differential rate corresponding to, say, a 45 Hz wave, while the trailing edge could give a voltage differential rate corresponding to 55 Hz. Thus, instead of "seeing" a 50 Hz waveform, the detection circuitry would see a low-level 45 Hz signal and a low level 55 Hz signal. This is precisely why phase relationships are so critical in electronic transmission.

It is also why substations must be set up every so far apart in transmission lines. The inductance of the lines, even with the attention paid to reactance, will become so great after so far that the transformers will not respond well to the now-misshapen waveforms. In addition to increasing the voltage to allow for line voltage drops, substations also reshape the waveform so it again closely approximates a true sine wave (normally through passive filtering here).

What matters is how much the frequency of the ~50 Hz signal can be expected to vary and the corresponding accuracy required to measure changes in it. The 50 Hz signal is not sending any information, it is just a sinusoidal, therefore it is inaccurate to state a low 'bandwidth' (I'm not even sure this is the right word to use in this context) means less information - it isn't carrying any in the first place.

If the frequency is varying with respect to time, that variance is information being sent and can be described by the bandwidth... the difference between the highest and lowest frequencies used. It's no different in theory from FM radio, just on a much more exacting scale (a scale I still maintain is impossible to accurately interpret outside of a laboratory).

A 50 Hz hum isn't constant though. It's only there if equipment generating a hum is present. It would only be cancelled if the noise cancelling technology is designed to filter ambient sound and is adaptive.

That's what noise-cancelling circuitry does.

Cars are only useful if they move from one point to another?

TheRedneck

Originally posted by TheRedneck
The difference in time between identical wave points on a 49.99 Hz and a 50.01 Hz (.02 Hz bandwidth) is approximately 0.000008 seconds. In order to detect this accurately would require a sampling rate of 125kHz. Typical sampling rates are 44.1 kHz, about one-third of the minimum rate required to make a differential measurement based on wavelength and about one-sixth the minimum rate to give reliable results..Thus the audio track cannot be used to provide a peak-to peak, trough-to-trough, or zero-crossing analysis.

The only other way to measure frequency, and the one commonly used, is to compare voltage differentials with respect to time at the zero-crossing points of the wave. This works with sinusoidal wave patterns fairly well, since the maximum differential voltage rates can be predicted from the frequency. That becomes impossible when phase shift is introduced, because the waveform is distorted from a true sine wave and the voltage differentials become impossible to predict without exact knowledge of the waveform itself.

The sampling rate you mention is required to measure a single cycle, not to measure the frequency of say, 200 cycles. Count 200 zero crosses and measure the time between the first and last.

reply to post by EasyPleaseMe

No. The sampling rate is the rate at which samples of the analog signal are taken and converted into digital data. In order to produce an analog signal (sound) from the digital recording, the sampling rate must be considerably higher than the frequency being sampled. In order to detect the frequency of the digital signal, there must be enough data available to determine which sample is indicative of which wave.

Example: If I tried to sample a 44.1 kHz signal using a 44.1 kHz sampling rate, I would get a flat line indicating no oscillatory signal whatsoever. The reason for this is that each time the signal is sampled, the wavefront is at the exact same position. If I try to sample a 22.05 kHz signal using a 44.1 kHz sampling rate, I could get several different results depending on where the wavefront was when the first sample was taken. If it was at 0°, I would still get a flat line because sine 0° = sine 180° = sine 360°. If it was at 45°, I would get samples corresponding to sine 45°, sine 225°, sine 45° and so on, which would look like a 22.05 kHz signal at 0.707 times the original signal. If the wavefront was at 90°, the resulting sampling would indicate a 22.05 kHz signal at full volume.

That's with a perfect harmonic of the sampling rate, which will never happen in reality. If the frequency being sampled is not a perfect harmonic and is not oscillating far slower than the sampling rate, the result is erroneous data that will not give a reasonable recreation of the original waveform. If the recreated waveform is not reasonably accurate, the frequency of that waveform is questionable at best and totally inaccurate at worst.

The above does not include the number of discrete intervals used in sampling, intervals which act to approximate the levels sampled. This further distorts the wave characteristics.

When the sampling rate is sufficiently higher than the frequency being sampled, each wave period is sampled several times, leading to a more accurate recording of the actual wave. In this case, the errors described above become irrelevant and the quality of the signal is maintained.

If I take a photo of a certain stretch of highway every five minutes, I can count the number of cars in each photo and get an approximation of how many cars are traveling that stretch of highway for every hour in the day. I can say with some level of certainty that there are, say, twice as many cars on average between 3:00 PM and 4:00 PM on a weekday as there are between 2:00 PM and 3:00 PM on a weekday. I cannot say with certainty how many cars were traveling that highway at 3:56:42 on August 29th 2013 (assuming that does not coincide with a photo), because the number of cars there is constantly changing and I only have limited data to go on. I can only make approximations about periods much longer than five minutes based on averages.

Same principle.

TheRedneck

Originally posted by TheRedneck
reply to post by EasyPleaseMe

No. The sampling rate is the rate at which samples of the analog signal are taken and converted into digital data. In order to produce an analog signal (sound) from the digital recording, the sampling rate must be considerably higher than the frequency being sampled. In order to detect the frequency of the digital signal, there must be enough data available to determine which sample is indicative of which wave.

Example: If I tried to sample a 44.1 kHz signal using a 44.1 kHz sampling rate, I would get a flat line indicating no oscillatory signal whatsoever. The reason for this is that each time the signal is sampled, the wavefront is at the exact same position. If I try to sample a 22.05 kHz signal using a 44.1 kHz sampling rate, I could get several different results depending on where the wavefront was when the first sample was taken. If it was at 0°, I would still get a flat line because sine 0° = sine 180° = sine 360°. If it was at 45°, I would get samples corresponding to sine 45°, sine 225°, sine 45° and so on, which would look like a 22.05 kHz signal at 0.707 times the original signal. If the wavefront was at 90°, the resulting sampling would indicate a 22.05 kHz signal at full volume.

That's with a perfect harmonic of the sampling rate, which will never happen in reality. If the frequency being sampled is not a perfect harmonic and is not oscillating far slower than the sampling rate, the result is erroneous data that will not give a reasonable recreation of the original waveform. If the recreated waveform is not reasonably accurate, the frequency of that waveform is questionable at best and totally inaccurate at worst.

The above does not include the number of discrete intervals used in sampling, intervals which act to approximate the levels sampled. This further distorts the wave characteristics.

When the sampling rate is sufficiently higher than the frequency being sampled, each wave period is sampled several times, leading to a more accurate recording of the actual wave. In this case, the errors described above become irrelevant and the quality of the signal is maintained.

Useful information for others but I am well versed in the Nyquist theorem and A/D systems in general thanks...

Originally posted by TheRedneck
If I take a photo of a certain stretch of highway every five minutes, I can count the number of cars in each photo and get an approximation of how many cars are traveling that stretch of highway for every hour in the day. I can say with some level of certainty that there are, say, twice as many cars on average between 3:00 PM and 4:00 PM on a weekday as there are between 2:00 PM and 3:00 PM on a weekday. I cannot say with certainty how many cars were traveling that highway at 3:56:42 on August 29th 2013 (assuming that does not coincide with a photo), because the number of cars there is constantly changing and I only have limited data to go on. I can only make approximations about periods much longer than five minutes based on averages.

Same principle.

TheRedneck

I am talking about a sliding temporal average with a set length of say a few seconds. Only a trend of the frequency change needs to be made over the duration of the recording because the mains frequency doesn't change at high speed. It can be seen in the video that they are using a trend with the recorded mains frequency having significantly more high frequency components.

reply to post by TheRedneck

When the sampling rate is sufficiently higher than the frequency being sampled, each wave period is sampled several times, leading to a more accurate recording of the actual wave. In this case, the errors described above become irrelevant and the quality of the signal is maintained.

As long as the sampling frequency is at least twice the frequency of the input signal, then the difference between 49.99 Hz and 50.01 Hz can easily be found, the only other necessitating requirements are that the window length is long enough and the amplitude isn't truncated too much via amplitude quantisation. Therefore, even at a sampling rate of for example, 125 Hz, after a window of several seconds the difference between 49.99 Hz and 50.01 Hz signals will have diverged significantly. There is no reason why Fourier analysis could not distinguish between two signals differing by 0.02 Hz.

Plenty of studies have been done on doing just that, here is another. In this study 1000 signals with a frequency randomly selected between 59.0 Hz and 61.0 Hz were sampled at a frequency of 1200 Hz. With 200 samples and a DFT of 20,000 points, the mean error in frequency was 0.0024% which corresponds to ~0.0009 Hz of error. Typical fluctuations in the electric power grid are between 0.005 Hz and 0.1 Hz. The paper then discusses phase discontinuities to find audio editing.

Again, 44.1 kHz is massive overkill to detect small changes in the frequency due to the reasons I list above. This different paper outlines exactly how analysis is done:

Signal decimation – Many digital audio recordings are recorded at high sampling frequencies – e.g., 44100 Hz. To detect the ENF, which is approximately 50 Hz, much lower sampling frequencies are allowed. The audio file is thus decimated to a sampling frequency of 300 Hz, which significantly reduces computational time.

Band pass filtering – The frequencies of interest are around 50 Hz, so the decimated audio file is
digitally band pass filtered from 49.5 Hz to 50.5 Hz to isolate the ENF.

Short Time Fourier Transform (STFT) – In discrete time STFT analysis, a signal is divided into J partly overlapping frames (figure 3) for which, after windowing and zero-padding, the frequency spectrum is calculated via a Discrete Fourier Transform (DFT). The jump H (in samples) between frames determines the time resolution of the final ENF pattern, while the amount of overlap M − H affects its smoothness. In our specific case, we have chosen H = 300 so that the extracted ENF pattern time resolution equals that of the database – i.e., 1 second. Each frame was windowed with a rectangular window and zero-padded by a factor of 4.

Peak frequency estimation – For each frequency spectrum, the frequency with maximum amplitude is estimated. As it is unlikely that this ‘peak frequency’ coincides exactly with a DFT frequency bin, quadratic interpolation around the bin with maximum amplitude is performed. The estimated peak frequency is stored as the ENF value for the corresponding frame, so that we end up with an extracted ENF pattern of J ENF values.

www.forensic.to...

It isn't just the UK that is doing this, it is Bavaria as well:

Is responsible for the fluctuation of the frequency of the power consumption. While all power generators operate largely constant at a frequency of 50 Hz, ie 50 cycles per second. But every time, for example, a factory is booted, the generators are heavily loaded - to the power plants to offset the increased demand after a short time. They increase the power, the rotors rotate faster again. And so the system frequency varies in a tiny range from 49.95 to 50.05 hertz - enough for the state police. Officials register the slightest fluctuations, and extends to them a small amount of power. The noise can be heard even when the recording device does not depend on the electrical network, for example when the film is taken with a portable video camera. Often enough, the sound emanating from other electrical equipment in the room to measure the variations in frequency can.

Since July 2010, officials cut the mains frequency, 24 hours a day, is a target database. Across Germany, the Bavarian LKA is a pioneer - this technique is relatively simple. The mains frequency could be measured at each outlet, says Dagmar defeated. "We in Munich have simply recognized as the first genius of this application."

(Translated by Google Translate)

www.sueddeutsche.de...

... continued

I am puzzled as to why you have ignored all prior discussion in this thread and instead you have insisted that those are the only ways to measure frequency, even though the discrete time Fourier transform is the fundamental basis for all DSP. The frequency measurement methods you have talked about seem like the ones you would use using cursors on a CRO.

And, I am still puzzled as to why you suggest phase shift due to line inductance or admittance would make frequency estimation difficult. There will be a phase shift depending on the length and frequency of the line, as well as a standing wave, but the frequency throughout a transmission line is exactly the same - and changing the phase angle doesn't matter. The only way I could see this happening is if there was harmonic distortion and you were counting the zero-crossing points, but that's what a band-pass filter is for.

That's what noise-cancelling circuitry does.

As I have already stated it depends on the specific circuitry.

reply to post by Arbitrageur


Correct.

Analysis of power line behavior has been a major technical subject of national intelligence agencies for many decades. One key area is attempting to determine facts regarding operation of machinery in a denied building/plant by measuring properties of the electrical power signal. One application for instance is large cascades of rotating machinery, such as centrifuges.

Application to forensic watermarking is a byproduct of R&D into the underlying technology.

reply to post by C0bzz

I am puzzled as to why you have ignored all prior discussion in this thread and instead you have insisted that those are the only ways to measure frequency, even though the discrete time Fourier transform is the fundamental basis for all DSP.

I'll try this one more time...

If you have an accurate representation of a signal, everything you say is possible. But digital sampling does not always give an accurate representation of a signal. In order to convert a signal from analog (the microphone) to digital, that signal must be sampled at periodic intervals, the most common being 44.1 kHz. All information between these sampling intervals is lost, gone kaput. All that is left are the voltage readings at these intervals. In addition, the voltage readings themselves are not completely accurate. If I am sampling, as an example, a voltage that I expect to range from 1 to 1.024 volts and I use 10 bits to record that data, I can be assured of an accuracy of 1 mV. Any information less than half that is lost. If the voltage is .0454632 V, I will read .045 V. The .0004632 V difference is lost, gone, kaput.

If the sampling rate is high enough compared to the frequency being recorded, and if enough bits are used to accurately represent the signal I am trying to record, the lost information becomes irrelevant.... interpolation when the signal is turned back into sound fills in the gaps, if you will, and reproduces a signal that is so close to the original as to be essentially identical for the purpose of reproduction of the information it contained (speech, music, etc.)

If the frequency gets too high, the sampling rate is not sufficient to retain enough information to accurately describe the wave. If the signal level gets too low, the quantization of the signal loses too much information to allow the wave to accurately be represented. Digital recording is not superior in accuracy to analog recording; just the opposite. The reason digital is preferred is because the information once recorded is not subject to time degradation and interference like analog. It can also be stored digitally, which is more efficient than audio recording, because the entire wave need not be stored. Only the samples of the wave need be recorded, and compression algorithms, like Fourier transforms, can be used to further decrease the size of the data.

Anyone posting technical observations should already know this.

All I am saying is that the recording equipment that would be used is not laboratory equipment; it is going to be audio/video recording equipment designed for amateur use, which means it is designed to record 20-20,000 Hz signals (human hearing) at a maximum... quite a lot of equipment uses something closer to 50-15,000 Hz, since that covers 90% of human hearing. It records at a signal level that will represent the normal range of sound, using sufficient bits to represent the information in that range.

Noise cancelling circuitry does vary depending on the design, but the entire concept is to reduce noise, noise being defined as continuous tones carrying no relevant information and feedback. That means any noise cancelling circuitry will at least attempt to remove a continuous tone. So if a 50 Hz hum is detected, it will be attenuated as far as the circuitry can attenuate it.

So what we have here is a signal which is specified to be below the level of human hearing and which therefore carries no relevant information pertaining to the purpose of the recording device, and which is continuous to all but the most advanced forensics techniques and is therefore considered noise and subject to attenuation by the noise cancelling capability of almost very modern device known, being recorded digitally (quantized and sampled) in a device designed to only record normal sound levels, and somehow, miraculously, the information which was never recorded magically reappears in its full glory accurate to such minute detain as to be capable of being compared to fluctuations in frequency so precise as to be beyond the capability of most laboratories.

OK, I give up. I cannot debate against magic and super powers of that magnitude. You win.

I only wish i had waited about building my shop so it could be magic too. I have to work within those silly old-fashioned laws of physics that keep restricting me from doing miraculous things.

TheRedneck

reply to post by TheRedneck


----

You would EXPECT that a government or power-supply agency would have SOME money
to at least BUY a decent sound card:

See /Sound-Blaster-X-Fi-Titanium-Pro-Audio-PCIe
www.soundblaster.com...

...that can record at 24bits per sample at 192 kHz so that obtaining sub-samples
of other waveforms INCLUDING mains hum (50/60hz) AND allow for general
DSP/FFT manipulation plus electrical noise waveform detection and resampling.

And since this card is less than 300 Euros/$500 U.S. I would expect the forensics
analysts could buy MORE THAN ONE of these recording devices and use it for
professional-level acoustic waveform detection and modeling thus making
the AVERAGE CONSUMER an "easy target" for unfair invasion of privacy.

So while this thread has sidetracked into the TECHNICAL aspects of
audio and main hum forensics, the larger picture is BEING IGNORED!
And that is that consumers are being UNFAIRLY deprived of basic
expectations of privacy. Technology such as this is making a MOCKERY
of the very term "Privacy" where governments, businesses, criminals
and other interested parties can use computers and science to literally
peer behind solid walls at will, without repercussion and THAT scares me!

And what's even more scary is on a basic level, almost NOTHING can be
done about it without MASSIVE and (and likely) VIOLENT revolution to
PURGE those who wish to abuse these technical superpowers.

C0bzz
I am puzzled as to why you have ignored all prior discussion in this thread and instead you have insisted that those are the only ways to measure frequency, even though the discrete time Fourier transform is the fundamental basis for all DSP.

If you're looking to track phase and frequency offset around a known tone and that's it, there are alternates to the DFT.

Just a few clicks on google and ignorance is denied for the day, whew, time for a beer! The 2nd paper explicitly mentions forensic applications.

www.ece.sunysb.edu...

www.bgu.ac.il...