Frequency Sweep Sonar System

I am conducting research in the implementation of algorithms for a 
frequency-sweep sonar system, and I have a few questions regarding the 
approaches to this problem.

(1) I have two transducers that are mounted next to each other facing a 
flat target situated at a fixed distance _y_ away from the transducers.  
One of the transducers produces a frequency-swept pulse, whereas the 
other receives the pulse when the first transducer is still 
transmitting.  The idea here is similar to FMCW radar.  Let the pulse 
sent out from the transducer be s[t], and the reflected pulse be s'[t].  
How do I digitally homodyne the two pulses so that a "peak" will occur 
in the frequency spectrum corresponding to a beat frequency that is 
associated with the distance _y_?  In FMCW radar, I believe that 
homodyning is performed by analog mixing and low-pass filtering the 
signal before ADC conversion.  I believe that this is done to ensure 
that the frequency difference is small enough so that s'[t] can be 
properly sampled at above Nyquist at a low enough sampling frequency.  
However, for the signal being sent out from the transducer a sampling 
frequency f_s = 96 kHz is more than adequate for my application.  How do 
I homodyne s[t] and s'[t] simply by the use of DSP?

(2) Given that the transducer sending out s[t] has a non-flat frequency 
response, how do I collect s[t] so that it can be digitally homodyned 
with s'[t]?  Do I want to digitally produce s[t] by the use of a 
mathematical model of the transducer?

(3) How do I set the cutoff frequency for the low-pass filter after s[t] 
and s'[t] are homodyned?  In the same manner as FMCW radar, I am 
interested in the beat frequency that is associated with distance.

(4) Does the wave s[t] produced by the sending transducer travel 
directly to the receiving transducer without contacting the target?  If 
so, how do I remove this direct wave from the signal?

(5) How do I select the separation distance between the two transducers? 

Thank you ever so much for all of your help!  

   

On 13 Feb, 22:07, Nicholas Kinar <n.ki...@usask.ca> wrote:
> I am conducting research in the implementation of algorithms for a
> frequency-sweep sonar system, and I have a few questions regarding the
> approaches to this problem.
>
> (1) I have two transducers that are mounted next to each other facing a
> flat target situated at a fixed distance _y_ away from the transducers.  
> One of the transducers produces a frequency-swept pulse, whereas the
> other receives the pulse when the first transducer is still
> transmitting.  The idea here is similar to FMCW radar.

The one parameter you need to address before doing anything else,
is source-reciever cross-talk. I don't know how cross-talk is handled
in FMCW radar, or even if it is a problem at all in that application.
I can, however, guarantee that it will be a problem with the sonar.

Do some simulation tests with radar simulators and see if the
algorithm can handle crosstalk ranging from 5% to 500% of the
power (intensity) of the recieved signals. If the radar simulator
can't
handle that, forget about using this processing scheme in sonars.

If the processing scheme surives the crosstalk test, the next test
is propagation delays. Remember that the speed of sound is very
low compared to radar (1500 m/s vs 3e8 m/s), which means that
there will be a response delay if the target is at any distance.
Again, play with the radar simulator and try to see how well it
works for detecting targets at long distaces, i.e. distances
on the scale of the pulse propagating for several sweep periods.

Be aware that radar processing technologies can not just be
transferred to sonar applications. The theory of Synthetic
Aperture Sonar (SAS) was published 30 years ago. The last time
I checked (the summer of 2000) a SAS demonstrator system
(the first ever?) had just been assembled and tested at sea
under very idealized circumstances. The problem was that
the speed of sound in water is so slow and the wavelengths
of the sound signals so short that the sonar platform
has to be kept still, on the order of a couple of millimeters,
for seconds. At sea, this is just impossible to achieve. I would
be very surprised if there are more than a couple of SAS
prototype systems in existence, world-wide.

Rune

Nicholas Kinar wrote:

> I am conducting research in the implementation of algorithms for a 
> frequency-sweep sonar system, and I have a few questions regarding the 
> approaches to this problem.
> 
[....]


> (1) I have two transducers that are mounted next to each other facing a 
> flat target situated at a fixed distance _y_ away from the transducers.  
> One of the transducers produces a frequency-swept pulse, whereas the 
> other receives the pulse when the first transducer is still 
> transmitting.  The idea here is similar to FMCW radar.  Let the pulse 
> sent out from the transducer be s[t], and the reflected pulse be s'[t].  
> How do I digitally homodyne the two pulses so that a "peak" will occur 
> in the frequency spectrum corresponding to a beat frequency that is 
> associated with the distance _y_?

This approach can work indeed. I tried it with the ultrasoind in the air.


   In FMCW radar, I believe that
> homodyning is performed by analog mixing and low-pass filtering the 
> signal before ADC conversion.  I believe that this is done to ensure 
> that the frequency difference is small enough so that s'[t] can be 
> properly sampled at above Nyquist at a low enough sampling frequency.

This is done for two reasons: the simplicity and the good dynamic range 
of the analog mixers.

> However, for the signal being sent out from the transducer a sampling 
> frequency f_s = 96 kHz is more than adequate for my application.  How do 
> I homodyne s[t] and s'[t] simply by the use of DSP?

Yes you can do it that way however the dynamic range of ADC may be a 
problem.

> (2) Given that the transducer sending out s[t] has a non-flat frequency 
> response, how do I collect s[t] so that it can be digitally homodyned 
> with s'[t]?  Do I want to digitally produce s[t] by the use of a 
> mathematical model of the transducer?

This is not going to be easy. The phase response is not flat also. The 
worst thing is the response of the transducer varies with the 
temperature and environment. You should use the signal narrow band 
enough to neglect those dependencies.


> (3) How do I set the cutoff frequency for the low-pass filter after s[t] 
> and s'[t] are homodyned?  In the same manner as FMCW radar, I am 
> interested in the beat frequency that is associated with distance.

You answered your own question.

> (4) Does the wave s[t] produced by the sending transducer travel 
> directly to the receiving transducer without contacting the target?

Yes, and this is a very bad effect which limits your dynamic range, and, 
hence, the operating distance.

   If
> so, how do I remove this direct wave from the signal?

Proper design of the transducers and their placement.

> (5) How do I select the separation distance between the two transducers? 

The bigger - the better.

Vladimir Vassilevsky

DSP and Mixed Signal Design Consultant

http://www.abvolt.com

Rune, thank you for these suggestions.  I will try the radar simulation 
tests and see if this can be done.

Once again, thank you!  This reply is greatly appreciated!

Nicholas

> 
> The one parameter you need to address before doing anything else,
> is source-reciever cross-talk. I don't know how cross-talk is handled
> in FMCW radar, or even if it is a problem at all in that application.
> I can, however, guarantee that it will be a problem with the sonar.
> 
> Do some simulation tests with radar simulators and see if the
> algorithm can handle crosstalk ranging from 5% to 500% of the
> power (intensity) of the recieved signals. If the radar simulator
> can't
> handle that, forget about using this processing scheme in sonars.
> 
> If the processing scheme surives the crosstalk test, the next test
> is propagation delays. Remember that the speed of sound is very
> low compared to radar (1500 m/s vs 3e8 m/s), which means that
> there will be a response delay if the target is at any distance.
> Again, play with the radar simulator and try to see how well it
> works for detecting targets at long distaces, i.e. distances
> on the scale of the pulse propagating for several sweep periods.
> 
> Be aware that radar processing technologies can not just be
> transferred to sonar applications. The theory of Synthetic
> Aperture Sonar (SAS) was published 30 years ago. The last time
> I checked (the summer of 2000) a SAS demonstrator system
> (the first ever?) had just been assembled and tested at sea
> under very idealized circumstances. The problem was that
> the speed of sound in water is so slow and the wavelengths
> of the sound signals so short that the sonar platform
> has to be kept still, on the order of a couple of millimeters,
> for seconds. At sea, this is just impossible to achieve. I would
> be very surprised if there are more than a couple of SAS
> prototype systems in existence, world-wide.
> 
> Rune
> 
> 

In article <OLqAh.63043$wc5.36820@newssvr25.news.prodigy.net>, 
antispam_bogus@hotmail.com says...

> 
> This approach can work indeed. I tried it with the ultrasoind in the air.
> 

Vladimir, this is very interesting.  How did you manage to homodyne the 
original and reflected waves?  Did you do this by an analog circuit?  Or 
did you use DSP?  Maybe I am going to answer my own question, but would 
you use point-by-point multiplication in the time domain to homodyne the 
two signals by DSP?


> This is not going to be easy. The phase response is not flat also. The 
> worst thing is the response of the transducer varies with the 
> temperature and environment. You should use the signal narrow band 
> enough to neglect those dependencies.
> 

As far as I know, the accuracy of resolving distance to a target is 
affected by the bandwidth with FMCW techniques, so this could be a 
limiting factor.  I will have to see if I can narrow the bandwidth.

Thank you ever so much for your reply!

Nicholas

Nicholas Kinar wrote:

>>This approach can work indeed. I tried it with the ultrasoind in the air.
>>
> Vladimir, this is very interesting.  How did you manage to homodyne the 
> original and reflected waves?  Did you do this by an analog circuit?  Or 
> did you use DSP?
>  Maybe I am going to answer my own question, but would 
> you use point-by-point multiplication in the time domain to homodyne the 
> two signals by DSP?

It was a pure analog solution. In our days, I would probably have done 
it with the microcontroller in the way similar to what you described.


>>This is not going to be easy. The phase response is not flat also. The 
>>worst thing is the response of the transducer varies with the 
>>temperature and environment. You should use the signal narrow band 
>>enough to neglect those dependencies.
>>
> 
> As far as I know, the accuracy of resolving distance to a target is 
> affected by the bandwidth with FMCW techniques, so this could be a 
> limiting factor. 

The worst problem is the loss of SNR due to the transmitter-receiver 
crosstalk, and due to the noise and incoherency of the sweep generator.

> I will have to see if I can narrow the bandwidth.

The transducers that were available had the relatively low operating 
bandwidth by themselves.


Vladimir Vassilevsky

DSP and Mixed Signal Design Consultant

http://www.abvolt.com

In article <owuAh.8044$gj4.2569@newssvr14.news.prodigy.net>, 
antispam_bogus@hotmail.com says...
 
Vladimir,  thank you once again for your reply.

> The worst problem is the loss of SNR due to the transmitter-receiver 
> crosstalk, and due to the noise and incoherency of the sweep generator.

I am going to try and remove crosstalk by a non-linear estimation method 
involving the minimization of a weighting function.  There is no 
guarantee that the converged solution will be stable, but proper 
transducer design does help to eliminate transmitter-receiver crosstalk.

>The transducers that were available had the relatively low operating 
>bandwidth by themselves

The transducers that I am using are similar to loudspeakers.  The 
response is not flat over the frequency range, but the bandwidth is 
still greater than ultrasonic transducers. 

Nicholas     

"Nicholas Kinar" <n.kinar@usask.ca> wrote in message 
news:MPG.203be4a264d2fc4d989680@news.usask.ca...
>I am conducting research in the implementation of algorithms for a
> frequency-sweep sonar system, and I have a few questions regarding the
> approaches to this problem.
>
> (1) I have two transducers that are mounted next to each other facing a
> flat target situated at a fixed distance _y_ away from the transducers.
> One of the transducers produces a frequency-swept pulse, whereas the
> other receives the pulse when the first transducer is still
> transmitting.  The idea here is similar to FMCW radar.  Let the pulse
> sent out from the transducer be s[t], and the reflected pulse be s'[t].
> How do I digitally homodyne the two pulses so that a "peak" will occur
> in the frequency spectrum corresponding to a beat frequency that is
> associated with the distance _y_?  In FMCW radar, I believe that
> homodyning is performed by analog mixing and low-pass filtering the
> signal before ADC conversion.  I believe that this is done to ensure
> that the frequency difference is small enough so that s'[t] can be
> properly sampled at above Nyquist at a low enough sampling frequency.
> However, for the signal being sent out from the transducer a sampling
> frequency f_s = 96 kHz is more than adequate for my application.  How do
> I homodyne s[t] and s'[t] simply by the use of DSP?
>
> (2) Given that the transducer sending out s[t] has a non-flat frequency
> response, how do I collect s[t] so that it can be digitally homodyned
> with s'[t]?  Do I want to digitally produce s[t] by the use of a
> mathematical model of the transducer?
>
> (3) How do I set the cutoff frequency for the low-pass filter after s[t]
> and s'[t] are homodyned?  In the same manner as FMCW radar, I am
> interested in the beat frequency that is associated with distance.
>
> (4) Does the wave s[t] produced by the sending transducer travel
> directly to the receiving transducer without contacting the target?  If
> so, how do I remove this direct wave from the signal?
>
> (5) How do I select the separation distance between the two transducers?
>
> Thank you ever so much for all of your help!

Nicholas,

A couple of comments:

Some forms of FMCW radar *use* inter-antenna crossfeed as a reference signal 
in the receiver - in order to compare the frequencies received.  The 
frequency difference is a measure of range.

There's a very good reason that FMCW won't work for sonar although it may 
work in air.  Volume Reverberation!
At short range the receiver will be flooded with volume reverberation 
signal.  So, how to get the nice mixing of target and near-field cross 
coupling?  Very tough to do.  Likely not possible.

It is very common for pulsed CW sonar to be reverberation limited at the 
receiver for short range targets.  After a time the volume reverberation has 
decayed to a point where the receiver becomes noise-limited.  Of course, 
this is for a broadband receiver.  But, isn't that what you'd use for FMCW? 
If the receiver has a filter bank or FFT then you expect noise limited 
performance except around zero doppler.  If the target isn't moving then 
that's where it will be and you're back to being reverberation limited.  (By 
"limited", I mean the "noise" floor in the channel is determined by this 
component and determines the necessary signal level for detection).

Fred

Nicholas Kinar wrote:

>>The worst problem is the loss of SNR due to the transmitter-receiver 
>>crosstalk, and due to the noise and incoherency of the sweep generator.
> 
> 
> I am going to try and remove crosstalk by a non-linear estimation method 
> involving the minimization of a weighting function.

This is going to be of little help.
The maximum gain of the receiver is limited by running into clipping on 
the feedthrough from the transmitter. Besides that, the white noise of 
the transmitter will hit the receiver all over the bandwidth.

>  There is no 
> guarantee that the converged solution will be stable, but proper 
> transducer design does help to eliminate transmitter-receiver crosstalk.

Don't fall into the occult methods of optimization :)

Vladimir Vassilevsky

DSP and Mixed Signal Design Consultant

http://www.abvolt.com

> > 
> > I am going to try and remove crosstalk by a non-linear estimation method 
> > involving the minimization of a weighting function.
> 
> This is going to be of little help.
> The maximum gain of the receiver is limited by running into clipping on 
> the feedthrough from the transmitter. Besides that, the white noise of 
> the transmitter will hit the receiver all over the bandwidth.
> 
> >  There is no 
> > guarantee that the converged solution will be stable, but proper 
> > transducer design does help to eliminate transmitter-receiver crosstalk.
> 
> Don't fall into the occult methods of optimization :)


Hmmm.  I wonder what I could do in this circumstance.

Once again, thank you for your help, Vladimir!

Nicholas