On Sun, 6 Oct 2019 20:51:06 +0200, Piergiorgio Sartor
<piergiorgio.sartor.this.should.not.be.used@nexgo.REMOVETHIS.de>
wrote:
>Hi all,
>
>is it possible to know, without any prior,
>if an incoming signal has a Doppler effect?
>
>That's it, to know if the source is moving
>to (or away from) the detector without any
>knowledge of the frequency.
>
>I can imagine that if the source is not
>moving directly towards (or away) the
>detector, but with an angle, the Doppler
>effect changes hence it is detectable.
>
>How about a direct motion?
>
>bye,
>
>--
>
>piergiorgio
I'm coming in a little late, but this is more or less what my master's
thesis was about, many decades ago.
And some of what I found has already been discussed, sort-of, but I'll
offer the following in case it's useful:
My goal was to be able to estimate velocity and range of vehicles from
the audio at a single sensor, so no beamforming, just signal analysis
from a single sensor.
I essentially computed the cross-correlation of time-adjacent spectra
to see how much they were shifting relative to each other, in order to
make a delta-f vs t plot, and then estimate f vs t once the
non-shifted spectra is determined (since you don't initially know how
much it is shifted).
As has been mentioned, this can be made to work reasonably well with
highly featured spectra, especially something like a sinusoid, which
has unambigious spectral features. The more it starts to look like
noise, or less like a sinusoid, the less well that particular method
works. I made a ton of audio recording of aircraft, from small recip
propeller driven aircraft to B-1B bombers taking off. The sound from
a B-1B at takeoff power is pretty close to noise, and pretty far from
a sinusoid.
I could generate synthetic data from a geometric model, vary the
distance, speed, initial frequency, etc., etc., and if it was a simple
spectra like a tone or a few tones, the algorithm worked very well,
even with a lot of AWGN added. Once the emitted spectra got more
complicated than that, though, it became harder to pick the needed
features out fo the spectral cross-correlation and results got a lot
less reliable. As long as an airplane was just whistling when it
went, we had it down, but anything much more than that was difficult.
There's also some interesting stuff about Doppler behavior that was
exploitatble, such that it was possible to know when the vehicle had
passed it closest approach point, and from that you could get the
non-shifted spectrum, and then velocity estimates could be made.
Without knowing how to do it, my major prof kept insisting that I also
estimate range. At one point I spent a weekend set out to prove
that it was not possible and instead had found a way to do it, and
that worked well with synthetic data, too.
This was in the late 80's without a lof of data acquisition capability
or processing horsepower, so compared to what could be done today it
was a bit limited. I demonstrated very good, reliable results with
synthetic data, some hand-wavy iffy results with a few live samples of
propeller airplanes and loud cars (a recording of the Indy 500 that
year, isolating cars on the straight), but nothing to get too excited
about.
I think there are better algorithms available today, and sensors and
processing horsepower are certainly cheap now, so beamforming or MUSIC
or something like what was previously suggested with a small
microphone array would probably work a lot better.
Anyway, it was a lot of fun at the time.
Reply by Piergiorgio Sartor●October 10, 20192019-10-10
On 10/10/2019 15.08, lito844@gmail.com wrote:
[...]
>> The idea is simple. If an emitter is moving the Doppler seen by each receiver is angle_of_arrival * velocity/wavelength. If the angle of arrival to each receiver is distinct so too will be the Doppler. FDOA measures the cross correlation with respect to frequency between pairs of received signals.If there is no motion the correlation is maximum at 0 Hz. Otherwise it peaks at the differential Doppler frequency.
>
> My previous post is in error. It should say Doppler is given by
> cos(angle_of_arrival) * velocity/wavelength.
Thanks a lot, that's probably the solution.
Thanks again,
bye,
--
piergiorgio
Reply by ●October 10, 20192019-10-10
On Thursday, October 10, 2019 at 5:57:57 AM UTC-7, lit...@gmail.com wrote:
> On Wednesday, October 9, 2019 at 10:05:12 AM UTC-7, Piergiorgio Sartor wrote:
> > On 09/10/2019 04.55:
> > [...]
> > > Under many circumstances one could detect Doppler effect without prior signal information (except a notion of its RF and BW) by measuring FDOA (frequency difference of arrival) using at least two sensors placed at different locations or colocated with separation between antennas.
> >
> > Is this similar or close (or the same)
> > as beamforming?
> >
> > bye,
> >
> > --
> >
> > piergiorgio
>
> The idea is simple. If an emitter is moving the Doppler seen by each receiver is angle_of_arrival * velocity/wavelength. If the angle of arrival to each receiver is distinct so too will be the Doppler. FDOA measures the cross correlation with respect to frequency between pairs of received signals.If there is no motion the correlation is maximum at 0 Hz. Otherwise it peaks at the differential Doppler frequency.
My previous post is in error. It should say Doppler is given by
cos(angle_of_arrival) * velocity/wavelength.
Reply by ●October 10, 20192019-10-10
On Wednesday, October 9, 2019 at 10:05:12 AM UTC-7, Piergiorgio Sartor wrote:
> On 09/10/2019 04.55:
> [...]
> > Under many circumstances one could detect Doppler effect without prior signal information (except a notion of its RF and BW) by measuring FDOA (frequency difference of arrival) using at least two sensors placed at different locations or colocated with separation between antennas.
>
> Is this similar or close (or the same)
> as beamforming?
>
> bye,
>
> --
>
> piergiorgio
The idea is simple. If an emitter is moving the Doppler seen by each receiver is angle_of_arrival * velocity/wavelength. If the angle of arrival to each receiver is distinct so too will be the Doppler. FDOA measures the cross correlation with respect to frequency between pairs of received signals.If there is no motion the correlation is maximum at 0 Hz. Otherwise it peaks at the differential Doppler frequency.
Reply by Piergiorgio Sartor●October 9, 20192019-10-09
On 09/10/2019 04.55, lito844@gmail.com wrote:
[...]
> Under many circumstances one could detect Doppler effect without prior signal information (except a notion of its RF and BW) by measuring FDOA (frequency difference of arrival) using at least two sensors placed at different locations or colocated with separation between antennas.
Is this similar or close (or the same)
as beamforming?
bye,
--
piergiorgio
Reply by Ethan Fenn●October 9, 20192019-10-09
>
> Given that white noise has energy down to 0 Hz, the energy will be
> shifted up, so there will be some energy at any arbitrary low frequency.
> That said, I'm not certain if the energy distribution will still be
> flat. Perhaps someone more familiar with this can comment, both on
> whether measuring Doppler shift on white noise is possible and whether
> it would be feasible in the real world.
>
Interesting question... if I'm thinking about it the right way, pink noise will be unchanged by a Doppler shift, because it has the same amount of energy per octave (or per any smaller frequency ratio interval). White noise will still be white, but will get more powerful when shifted down in frequency and less powerful when shifted up. Brown noise stays brown but changes energy in the opposite way.
-Ethan
Reply by ●October 8, 20192019-10-08
On Sunday, October 6, 2019 at 11:55:36 AM UTC-7, Piergiorgio Sartor wrote:
> Hi all,
>
> is it possible to know, without any prior,
> if an incoming signal has a Doppler effect?
>
> That's it, to know if the source is moving
> to (or away from) the detector without any
> knowledge of the frequency.
>
> I can imagine that if the source is not
> moving directly towards (or away) the
> detector, but with an angle, the Doppler
> effect changes hence it is detectable.
>
> How about a direct motion?
>
> bye,
>
> --
>
> piergiorgio
Under many circumstances one could detect Doppler effect without prior signal information (except a notion of its RF and BW) by measuring FDOA (frequency difference of arrival) using at least two sensors placed at different locations or colocated with separation between antennas.
>> Given that white noise has energy down to 0 Hz, the energy will be
>> shifted up, so there will be some energy at any arbitrary low frequency.
>> That said, I'm not certain if the energy distribution will still be
>> flat. Perhaps someone more familiar with this can comment, both on
>That's is my conjecture, it will not
>be white, I just simplified.
>> whether measuring Doppler shift on white noise is possible and whether
>> it would be feasible in the real world.
>Yep, that's one question!
Depends on the white noise. Not all white noise is created equal.
Seems to me something like the "random telegraph process", which is
white but not Gaussian, could have its Doppler shift measured.
Steve
Reply by Piergiorgio Sartor●October 8, 20192019-10-08
On 07/10/2019 22.16, Phil Martel wrote:
[...]
> Given that white noise has energy down to 0 Hz, the energy will be
> shifted up, so there will be some energy at any arbitrary low frequency.
> That said, I'm not certain if the energy distribution will still be
> flat. Perhaps someone more familiar with this can comment, both on
That's is my conjecture, it will not
be white, I just simplified.
> whether measuring Doppler shift on white noise is possible and whether
> it would be feasible in the real world.
Yep, that's one question!
bye,
--
piergiorgio
Reply by Phil Martel●October 7, 20192019-10-07
On 10/7/2019 12:08, Piergiorgio Sartor wrote:
> On 07/10/2019 15.59, Phil Martel wrote:
>> On 10/6/2019 16:39, Piergiorgio Sartor wrote:
>>> On 06/10/2019 22.25, Steve Pope wrote:
>>>> Piergiorgio Sartor
>>>> <piergiorgio.sartor.this.should.not.be.used@nexgo.REMOVETHIS.de> wrote:
>>>>
>>>>> Hi all,
>>>>>
>>>>> is it possible to know, without any prior,
>>>>> if an incoming signal has a Doppler effect?
>>>>>
>>>>> That's it, to know if the source is moving
>>>>> to (or away from) the detector without any
>>>>> knowledge of the frequency.
>>>>>
>>>>> I can imagine that if the source is not
>>>>> moving directly towards (or away) the
>>>>> detector, but with an angle, the Doppler
>>>>> effect changes hence it is detectable.
>>>>>
>>>>> How about a direct motion?
>>>>
>>>> Not as you have phrased it.� Even if the relative
>>>> motion is at an angle, the component of motion
>>>> along the line connecting the source and the receiver
>>>> determines the Doppler shift, and is still constant.
>>>
>>> Why it is constant?
>>>
>>> If the motion does not go towards to (or away
>>> from) the receiver the relative speed is not
>>> constant, hence the Doppler should also not be.
>>>
>>> Let me try to portrait it:
>>>
>>> ------ source motion ------->
>>> a\��������� b|�������� c/
>>> �� \��������� |�������� /
>>> ��� \�������� |������� /
>>> ���� \��� (receiver)� /
>>>
>>> When on the left (a), there is relative
>>> fast motion, like on the right (c).
>>> When exactly above (b), in that moment, there
>>> is no motion, or minimal.
>>> At least in the horizontal direction (more or
>>> less, clearly)
>>>
>>> There should be a trigonometric (sin(a) or
>>> cos(a)) relationship, I guess.
>>>
>>> Am I missing something?
>>>
>>>> One needs to know something about the sgnal.
>>>
>>> Let's assume the signal is (more or less) white
>>> noise, could this help?
>> What does frequency shifted white noise look like?� If I had to guess,
>> I'd say white noise, so that is probably the hardest case.� Start with
>> a sinusoid.� The frequency change will be fairly obvious.
>
> I do not think the white noise will lead
> to white noise.
> If the shift is upward, there will be
> no low frequencies, i.e. not white anymore.
I disagree with the "no low frequencies". The shift is dependent on
frequency.
https://en.wikipedia.org/wiki/Doppler_effect
f=\left(1+{\frac {\Delta v}{c}}\right)f_{0}
Given that white noise has energy down to 0 Hz, the energy will be
shifted up, so there will be some energy at any arbitrary low frequency.
That said, I'm not certain if the energy distribution will still be
flat. Perhaps someone more familiar with this can comment, both on
whether measuring Doppler shift on white noise is possible and whether
it would be feasible in the real world.
>
> On the other hand, the white noise is
> "quasi white", since it is anyway low
> pass, hence a shift downward could be
> still detectable. It depends.
>
> In any case, I do not have control over
> the source, I can only speculate on it.
> One "speculation" is that it could be
> (quasi) white noise or similar.
>
> bye,
>