High Frequency Linear Array BeamForming

[nauman]

Hi all

I am doing some research on High Frequency Linear Array beam forming (for imaging sonar). For that purpose, a linear Rx Array of 100 channels have been constructed with inter channel spacing of 14.3mm. A transmitting Tx probe operating at 420 KHz (approx. 300 cycles CW pulse) is placed at distance of approx. 8.9m from Rx array in water tank of dimensions (Length=12m, Width=8m, Depth=8m). The depth of Tx probe and Rx array is around 4m. The array data has been simultaneously acquired at sampling rate of 2Msamples/s/channel using NI ADC cards. The ADC cards and pulse generator for Tx probe are accordingly synchronized.

I have done some array beam forming on Rx array data along with focusing (at 9.3m approx.) and beam pattern result is drastic (far worse than simulated one) and shown below. The useful sector is around 10 degree and MATLAB 'chebwin' shading with -21dB side lobes has been used.

If someone has done such work and if he/she is interested, i can share digitized raw data of Rx array in MATLAB '.mat' format so that he/she can beam form this data and share beam pattern results with me?Unfortunately, i was unable to attach zip file (containing .mat file) in this forum. I want to know in steps where is issue?, in my processing, in setup or in both?

Note: This post is in continuation of work posted earlier as "Linear array beam pattern" with some modifications of course.

[Mentor link to previous thread]: https://www.physicsforums.com/threads/linear-array-beam-pattern.969721/

Thanks & Regards
Nauman

 

[berkeman]

Summary:: Higher side lobes observed in Receive Array Beam Pattern in water Tank. (Continuation of work previously posted as 'Linear array beam pattern')

A transmitting Tx probe operating at 420 KHz (approx. 300 cycles CW pulse) is placed at distance of approx. 8.9m from Rx array in water tank of dimensions (Length=12m, Width=8m, Depth=8m). The depth of Tx probe and Rx array is around 4m. The array data has been simultaneously acquired at sampling rate of 2Msamples/s/channel using NI ADC cards.

What are the characteristics of your anti-alias filter(s)? Your 4x oversampling is not very high. It looks like you have a lot of aliasing going on in your processed data, IMO.

And are the ADC conversions at your NI ADC cards all running at the same time? So you have 100 NI ADC cards that you are somehow synchronizing the clocks on? Which NI ADC cards are you using?

 

[nauman]

I am using 07 NI PXIe-6368 Cards each having 16 ADCs with simultaneous sampling of 2 Msamples/s/channel. These cards are installed in PXIe-1078 chassis which synchronize them internally.

 

[berkeman]

I am using 07 NI PXIe-6368 Cards each having 16 ADCs with simultaneous sampling of 2 Msamples/s/channel. These cards are installed in PXIe-1078 chassis which synchronize them internally.

Impressive cards! But they appear to rely on external anti-alias filtering, which it sounds like you have not implemented yet. Since you are only 4x oversampling, you will need quite steep anti-alias LPFs for each of the 100 channels. If you were doing more like 10x or 20x oversampling, you might be able to get away with 2-pole passive filters. But at 4x oversampling, I think they will need to be active LPFs with more poles. I'm not sure which polynomial would be best to use for this, since you would like sharp skirts but also good matching between all of the channels.

I also don't know offhand if there is some DSP trick that you can use to ease the need for sharp anti-alias LPFs since you have so many adjacent and related channels.

 

[nauman]

Each acoustic channel has a preamplifier with Band pass filter centered at frequency (F) 420 KHz with 3dB attenuation for 404 KHz <F< 436 KHz.

 

[berkeman]

Each acoustic channel has a preamplifier with Band pass filter centered at frequency (F) 420 KHz with 3dB attenuation for 404 KHz <F< 436 KHz.

3dB/decade rolloff with a 4x oversampling rate? What does that say about aliasing?
Sorry, I had trouble translating your statement into a standard rolloff rate.

Also, have you tried calibrating your setup with a CW transmission from your transmitting device and data acquisitions at your 100 channels? You can hand-tune the receive intervals to try to avoid windowing issues.

 

[berkeman]

Also, is there such a thing as anechoic structures in such aquatic chambers? We commonly use them in RF and acoustic test chambers. It seems like it would be a good idea for you to attenuate reflections from the 5 hard surfaces of the pool (I'm not sure how to handle the surface of the water as a reflector...)

https://antennatestlab.com/wp-conte...-Anechoic-Chamber-Quad-Ridge-Horn-Antenna.jpg

 

[nauman]

3dB/decade rolloff with a 4x oversampling rate? What does that say about aliasing?

Should not this bandpass filter be enough to suppress high frequency components? I will monitor pre-amplifier's output on Oscilloscope/Spectrum analyzer to check whether there are higher frequency components (>1M Hz) present?

Also, have you tried calibrating your setup with a CW transmission from your transmitting device and data acquisitions at your 100 channels? You can hand-tune the receive intervals to try to avoid windowing issues.

I have calibrated my NI DAQ system using external Function Generator and all seems OK to me.

 

[nauman]

Also, is there such a thing as anechoic structures in such aquatic chambers?

I do not know about RF, but underwater acoustic tiles using for attenuation purpose are very costly .

 

[berkeman]

Should not this bandpass filter be enough to suppress high frequency components? I will monitor pre-amplifier's output on Oscilloscope/Spectrum analyzer to check whether there are higher frequency components (>1M Hz) present?

When you are digitizing a pulsed waveform, there will be higher frequency components. What windowing function are you using? What fraction of the pulsed waveform are you digitizing?

When you transmit a CW waveform from your transmitter, do the digitized time waveforms from your transducers look clean and have the expected phase shifts? When you do a THD analyses on those waveforms do you see harmonics many of 10s of dB down? When you use your spectrum analyzer on the received CW waveforms out of the preamps, do you see any distortions from reflections?

And yes, anechoic tiles are expensive. But there is an important reason that we use them in RF and audio anechoic and semi-anechoic chambers...

 

[nauman]

When you are digitizing a pulsed waveform, there will be higher frequency components. What windowing function are you using? What fraction of the pulsed waveform are you digitizing?

On transmission side, i am not using any window, transmitting rectangular CW Pulse of around 300 cycles. On reception side, i am acquiring around 33ms data of each sensor data and perform beam forming on this data using chebwin window shading with -21dB side lobes.

When you transmit a CW waveform from your transmitter, do the digitized time waveforms from your transducers look clean and have the expected phase shifts?

Yes, you are right. When monitoring individual channels, the phases of digitized time waveform of individual channels does not correspond to ones simulated. What can be the reason for it?.

 

[berkeman]

Yes, you are right. When monitoring individual channels, the phases of digitized time waveform of individual channels does not correspond to ones simulated. What can be the reason for it?.

Ah, that's a good clue. If all of the digitizing synchronization mechanisms are working, then you should be able to look at the individual digitized sine waves and see the phase moving slowly as you look farther and farther out from the center channel. But with only 4x oversampling, you may not get much of a sine wave to look at.

 

[berkeman]

BTW, do you have access to a 4-channel oscilloscope with GHz sampling that you can use to test your system? If so, I would connect Ch1 to the middle preamp/filter output, and Ch2 and Ch3 to the adjacent preamp/filter outputs. Then capture some Tx waveforms to be sure you see what you expect, and then step Ch2 and Ch3 out to each adjacent pair of inputs, and so on. That will at least prove out your basic Rx array system, and may show you any gross issues with reflection distortions in the pool.

Then you can acquire multiple o'scope datasets that are all triggered on Ch1 and run those through your processing algorithm to ensure that everything works with much higher oversampling (1GHz/433kHz). If you get more expected results with that analysis, that shows that you may need an even more costly data acquisition setup to get this to work.

Or, instead of more costly, it may just mean that you need to design and build your own custom data acq circuitry for this application. The card specs of 16 bits at 2MS/s seems mismatched for this application to me. Depending on your overall system specs, I would think that 12b at 100MS/s or 8b at 1GS/s would work much better for this application. If I were working on this project, I would consider designing a custom data acq board for all of these channels that included much better anti-alias filtering on all of the channels, and much higher synchronized S/H and digitizing frequencies.

Alternately, if you need to purchase off-the-shelf hardware (if you don't have EE design services available for your project), consider high-speed USB oscilloscopes like the PicoScope products:

https://www.picotech.com/products/oscilloscope

(Disclosure -- I am using a 4-channel PicoScope for a high performance data acq system right now at work and am very happy with the performance that I'm getting after one of our DSP savant engineers used the 1GHz oversampling on my 150kHz pulsed network communication packets)

I haven't tried to synchronize multiple PicoScope captures, but hopefully that would be do-able.

 

[marcusl]

I agree with the observations made in this thread and the previous one. Of the concerns raised, I think the top ones are

1. Reflections and standing waves in the tank
2. Antialiasing filter
3. Instability in your electronics and other parts of your apparatus (you mentioned huge randomly-varying phase shifts and no repeatability from pulse to pulse)

You need to fix all of these.
BTW, how rigid is your array and how firmly are your antennas mounted in the tank? Variations and flatness should be held to better than one-tenth wavelength, which is 400 microns. (For example, the ends of your array that are 1.4 m apart must remain flat and rigid with respect to each other to within 400 microns.)

 

[nauman]

1. Instability in your electronics and other parts of your apparatus (you mentioned huge randomly-varying phase shifts and no repeatability from pulse to pulse)

Hi Marcusl
To check the NI DAQ system, i have used NI DAQ system and a Function Generator (FG) in external trigger mode. As soon as an external trigger pulse is received, FG outputs a CW pulse of 420 KHz with a programmed delay (approx. 6ms corresponding to 8.9m) to all channels of DAQ system simultaneously. At the same time, NI DAQ also starts digitization upon reception of external trigger pulse. Once 33ms data is acquired of all channels, i perform my processing and plot beam pattern (with Focusing OFF) which matches the simulated one. As per my understanding, this experiment simulates a scenario where Tx probe is placed at 0 deg w.r.t array center and all channels receive pulses without any phase difference among them.

I have also tried to do above experiment with 100 preamplifiers in loop also (i.e. giving signal to 100 preamplifiers connected with DAQ system simultaneously from FG) but FG was unable to drive 100 preamplifiers (each with 50 ohm input impedance) simultaneously perhaps due to very low net input impedance of 0.5 ohm .

Nevertheless, I have checked phases of preamplifier outputs w.r.t each other in pairs, they do have phase and gain differences among them but they do not vary with time. Time variation of Phase/Gains are observed in underwater setup especially if water is not still in Tank due some disturbances.

BTW, how rigid is your array and how firmly are your antennas mounted in the tank?

Acoustic modules are rigidly mounted on a water sealed array with preamplifiers housed in it.

1. Reflections and standing waves in the tank

Reflections from boundaries in water tank is the problem out of hand right now. Currently, I am using omni directional Tx probe for beam pattern measurement. However, efforts are in hand to acquire narrow directional Tx probe (with 3 dB beamwidth of 10 degree approx.). Should this help to avoid unwanted reflections from tank boundaries?

1. Antialiasing filter

I will try to measure phases as well as spectrum of pre-amplifier's output on Oscilloscope with G samples/s rate as suggested by berkeman.

 

[nauman]

BTW, do you have access to a 4-channel oscilloscope with GHz sampling that you can use to test your system? If so, I would connect Ch1 to the middle preamp/filter output, and Ch2 and Ch3 to the adjacent preamp/filter outputs. Then capture some Tx waveforms to be sure you see what you expect, and then step Ch2 and Ch3 out to each adjacent pair of inputs, and so on. That will at least prove out your basic Rx array system, and may show you any gross issues with reflection distortions in the pool.

Then you can acquire multiple o'scope datasets that are all triggered on Ch1 and run those through your processing algorithm to ensure that everything works with much higher oversampling (1GHz/433kHz). If you get more expected results with that analysis, that shows that you may need an even more costly data acquisition setup to get this to work.

I will try to measure phases as well as spectrum of pre-amplifier's output on Oscilloscope with G samples/s rate .

Or, instead of more costly, it may just mean that you need to design and build your own custom data acq circuitry for this application. The card specs of 16 bits at 2MS/s seems mismatched for this application to me. Depending on your overall system specs, I would think that 12b at 100MS/s or 8b at 1GS/s would work much better for this application.

As i am using frequency domain beam former, i thought 4 times sampling would be more than enough.

 

[nauman]

If so, I would connect Ch1 to the middle preamp/filter output, and Ch2 and Ch3 to the adjacent preamp/filter outputs. Then capture some Tx waveforms to be sure you see what you expect, and then step Ch2 and Ch3 out to each adjacent pair of inputs, and so on.

I have measure spectrum of randomly selected channels using Oscilloscope having sampling rate of G samples/s and did not find any significant higher frequency components (840 KHz signal peak was there about 60 dB below 420 KHz peak). Moreover, i found phase difference among randomly selected channels using in water setup on Oscilloscope and found them approximately same as using NI DAQ system. However, these phase are not as per theoretical values (significantly deviate from theoretical ones)

 

[berkeman]

However, these phase are not as per theoretical values (significantly deviate from theoretical ones)

In what ways? Do you have plots that you can share?

 

[berkeman]

BTW, it just occurred to me that ensuring that there is no interference between the adjacent RX array elements is important. If the support structure or each RX element is reflecting any energy that gets absorbed by adjacent RX elements, that also will interfere with the receive array performance.

How are these RX elements joined together? Is there any opportunity for reflections off of them or the support structure to interfere with adjacent/nearby elements?

That is why I would keep working with the GHz 'scope receive testing for now, to debug any of these kinds of issues (including pool wall reflections) before spending much more time on processing the Data Acq system data.

 

[Baluncore]

Each acoustic channel has a preamplifier with Band pass filter centered at frequency (F) 420 KHz with 3dB attenuation for 404 KHz <F< 436 KHz.

I would want to know the details of those filters and pre-amps so I could investigate the phase sensitivity to component tolerances.

The transmitted envelope of 420 kHz in a quiet environment should dominate all other sources. So why do you need an anti-alias filter when there is very little energy up there that might be down converted into the band of interest.

The received signal will be the vector sum of all received reflection phasors. I would expect that there will be a phase and/or amplitude change each time a more distant reflected phasor reaches the transducer and is added to the received signal. Maybe you could examine how the phase changes during the time record for one transducer. That should show how many different reflective paths are involved, and the travel distance via the different paths.

 

[Fred Wright]

I look at your problem from a post processing point of view. As a hobby project I have developed a maximum entropy noise reduction algorithm for image processing. I acquired the data from the image you posted using opencv and applied it to my c++ routine.

My algorithm is a modified version of the algorithm developed in the paper, J. Skilling and R. K. Bryant, "Maximum entropy image reconstruction: general algorithm", Mon. Not. R. astr. Soc. (1984) 211, 111-124. If you think it would be helpful for your research I can process some more of your data.

 

[nauman]

In what ways? Do you have plots that you can share?

With transmitting probe in front of array at 8.4m distance and focusing at receive end being done at 9.3m, the simulation results of Phase difference b/w reference sensor (i.e. 50) and rest of sensors is shown below:

The actual results received using same scenario in Tank are shown below:

Issue is that these phase differences vary with even slightly change in setup, e.g. Transmit probe position, relative depth etc.

 

[nauman]

If you think it would be helpful for your research I can process some more of your data.

Hi Fred Wright
Thanks. Do you mean raw unprocessed data of array sensors or just beam pattern plots? I am asking because if some one is kind enough to process the raw sensor data and share his/her results, i will be able to clarify at least my processing side.

 

[Baluncore]

Issue is that these phase differences vary with even slightly change in setup, e.g. Transmit probe position, relative depth etc.

That suggests you have a reflection from the water surface interfering with the direct path.
Change the surface and see if it changes the pattern. Maybe float a sponge or a shaggy carpet on the surface to destroy the perfect mirror.

 

[nauman]

I would want to know the details of those filters and pre-amps so I could investigate the phase sensitivity to component tolerances.

I do not have detailed design specifications of preamplifiers/filters, however generic specifications are as:

• Input referred noise < 3uV
• Source & Load Impedance = 50 ohm
• Gain variation = 0 to 80 dB
• Phase/Gain dispersion = +/- 3º/2dB

I have tested individual preamplifiers/filters w.r.t. reference preamplifier/filter (by applying same signal through FG to both reference and to be tested preamplifier/filter) and there are phase and gain differences but these are constant and do not change.

The received signal will be the vector sum of all received reflection phasors. I would expect that there will be a phase and/or amplitude change each time a more distant reflected phasor reaches the transducer and is added to the received signal.

As I have observed during testing that phases of preamplifiers vary among them, is it possible that different acoustic sensors also have phase variation among them (irrespective of phases due path differences)? And if yes, is it any way to measure these phase variations ?

 

[Baluncore]

If you are operating close to a transducer resonance you may get phase differences between transducers.

What is the make and model of the sonar receive transducers? Do you have a datasheet? What is their resonant frequency?

 

[nauman]

If you are operating close to a transducer resonance you may get phase differences between transducers.

What is the make and model of the sonar receive transducers? Do you have a datasheet? What is their resonant frequency?

I am using customized developed transducers with following receive sensitivity vs frequency graph:

 

[Baluncore]

300 cycles at 420 kHz;
420 kHz = 2.38 usec;
Speed of sound in water 1480 m/sec;
Wavelength is 3.52 mm;
2.38 usec * 300 = 714.28 usec;
Pulse envelope is 1.057 metre long;

Tx probe at distance of approx. 8.9m from Rx array in water.
tank of dimensions (Length=12m, Width=8m, Depth=8m).
The depth of Tx probe and Rx array is around 4m;

The length of the pulse train in water is about 1.057 metre.
The path is 8.9 m direct,
or reflected via bottom, side or surface is 4 metre away.
What is path length via 4 metre side or surface reflection ?
2 * Sqrt( 4.45^2 + 4^2 ) = 12.0 metre which is 3 metre further than the direct path

The tank is large enough to contain the 1.057 m long pulse without reflections from the walls or surface.

The sampled data needs to include a quiet period at each end.

On reception side, i am acquiring around 33ms data of each sensor data

You should be using less than 1 msec of data since after that, the multipath reflections will make a mess of the signal.
1. How do you identify the start of valid data? Maybe first arrival time at any sensor?
2. How do you identify the end of valid data? Maybe 1 msec after start?
3. How long is it before the first indirect reflection arrives? Maybe 1.5 msec.

 

[nauman]

300 cycles at 420 kHz;
420 kHz = 2.38 usec;
Speed of sound in water 1480 m/sec;
Wavelength is 3.52 mm;
2.38 usec * 300 = 714.28 usec;
Pulse envelope is 1.057 metre long;

Tx probe at distance of approx. 8.9m from Rx array in water.
tank of dimensions (Length=12m, Width=8m, Depth=8m).
The depth of Tx probe and Rx array is around 4m;

The length of the pulse train in water is about 1.057 metre.
The path is 8.9 m direct,
or reflected via bottom, side or surface is 4 metre away.
What is path length via 4 metre side or surface reflection ?
2 * Sqrt( 4.45^2 + 4^2 ) = 12.0 metre which is 3 metre further than the direct path

The tank is large enough to contain the 1.057 m long pulse without reflections from the walls or surface.
1. How do you identify the start of valid data? Maybe first arrival time at any sensor?
2. How do you identify the end of valid data? Maybe 1 msec after start?
3. How long is it before the first indirect reflection arrives? Maybe 1.5 msec.

The raw data of middle sensor (i.e. 50#) is as:

The zoom data of direct pulse and indirect pulse is as:

As sampling is being done at 2 Msamples/s, direct pulse start time is approx. 5.94ms (11887/2e6) which is around 8.85m using sound speed of 1500m/s. The indirect pulse start time is approx. 7.3ms (14635/2e6) which is around 10.95m. The time difference b/w both pulses is around 1.4ms approx.

The sampled data needs to include a quiet period at each end.
You should be using less than 1 msec of data since after that, the multipath reflections will make a mess of the signal.

Before doing beam forming, i only use direct pulse data (e.g. in above case samples from 11000 to 13500 only) and force other data to zero in software for all sensors.

 

[Baluncore]

OK, so you use only direct signals, and have eliminated all tank boundary reflections.

You will find the spurious phase variation in the last place you look, because then you can stop looking, but where should you look next?

1. The separation between the elements in the array is 14.3mm, which is much greater than the wavelength of 3.52 mm. That will lead to ambiguity of received phase. The main lobe will have a null at Atan( 3.52 / 14.3 ) = 13.82°;
The 100 element * 14.3 mm = 1430 mm long array will require a rigid mounting with alignment accurate to better than 0.3 mm at the ends of the array.

2. Coaxial transmission lines will look like capacitive lumps at 420 kHz. If lines are not all the same length there will be phase differences due to reactive transducer loading. Those should be steady repeatable errors.
Cables can be microphonic. Are cables subjected to 420 kHz in the water? Or do cable bundles drift, flexing in the water? That would produce variable phase errors.

3. Can you describe the physical connections for the channels. Where are the filters and pre-amplifiers located, and are all the interconnections the same length. I would look for hum loops where ground currents flow in signal cables, or through the water.

4. Is the water in the tank absolutely still, or is there circulation that could introduce Doppler shifts?

5. Is there a low frequency rumble such as a heavy motor, Air Con, or road noise present? That could position-modulate the phase at the transducers.

6. If you reduce the TX pulse by 20 dB, or 40 dB, does the spurious phase still occur? That will eliminate cross-modulation in RX transducers and processing electronics.

 

[nauman]

1. The separation between the elements in the array is 14.3mm, which is much greater than the wavelength of 3.52 mm. That will lead to ambiguity of received phase. The main lobe will have a null at Atan( 3.52 / 14.3 ) = 13.82°;
The 100 element * 14.3 mm = 1430 mm long array will require a rigid mounting with alignment accurate to better than 0.3 mm at the ends of the array.

3. Can you describe the physical connections for the channels. Where are the filters and pre-amplifiers located, and are all the interconnections the same length. I would look for hum loops where ground currents flow in signal cables, or through the water.

As separation is greater than (wavelength/2), 'grating lobes' occurs but beyond +/-6 deg. That's why only +/-5 deg sector has been used here. The sensors are mounted on a rigid water tight hollow cylinder outer surface with preamplifiers/filters fitted inside the hollow cylinder. There is approx. 0.5m coaxial cable b/w sensor and corresponding preamplifier. The two water sealed 18m long cable sets each carrying 50 preamplifiers/filters output goes to surface through two glands in hollow cylinder.

 

[Baluncore]

Changing the RX array position very slightly changes the RX phase differences significantly. That suggests there is cross-coupling between the signals somewhere other than in the water.
I can't see how to progress investigating signal integrity without all detailed circuit and mounting information. Such as;

The sensors are mounted on a rigid water tight hollow cylinder outer surface with preamplifiers/filters fitted inside the hollow cylinder.

What is the tube made from? Is it grounded? How and where are the signals and power supply grounded.
Or;

The two water sealed 18m long cable sets each carrying 50 preamplifiers/filters output goes to surface through two glands in hollow cylinder.

Are the 18 m long cables each 50 twisted pairs, or 50 coaxial cables. How are the two ends of those transmission lines terminated to prevent reflection? and what about cross-talk?

When you compare the phase of an element against the two adjacent elements, is the phase difference of the first few cycles the same as the difference for the last few cycles, or does phase vary during the RX pulse? Are some, or all, channels unreliable?