21cm band. Refurbishing the dish with a new Cantenna.

[SteveBz]

On 13/03/2022 at 02:10, ZiHao said:

Hello Steve,

The genuine rtlsdr just arrived few days ago, so you asked at the right time! I was using a clone rtlsdr which has a lot of noise before this, and the genuine one shows a lot of improvement. I am using a very simple python script with pyrtlsdr for data acquisition, this is similar to Victor's method. My setup is LNA-->5m LMR400-->Rtlsdr-->Rpi. Still working on a cleaner baseline for easier data processing and hopefully able to mount the dish on my equatorial mount soon.

 

Hi Zi Hao,

How did you get to be so beautifully clean and without noise spikes?  What software did you use?  Have you stacked several images?

Maybe I should put a wire fence around the scope like a chicken run!

Kind regards

Steve.

[ZiHao]

39 minutes ago, SteveBz said:

Hi Zi Hao,

How did you get to be so beautifully clean and without noise spikes?  What software did you use?  Have you stacked several images?

Maybe I should put a wire fence around the scope like a chicken run!

Kind regards

Steve.

Hello Steve,

When I was using the clone rtlsdr, I used to take a reference spectrum at another frequency, where there is no hydrogen line signals, and subtract that with the spectrum with hydrogen line signal. It should be clear that whether or not there's a signal, despite the clone dongle having a lot of noise spikes. With the genuine rtlsdr, I am trying to fit a polynomial curve to the source spectrum to avoid adding extra noise when subtracting. The result I posted is manually done by eyeballing the curve that fits the overall shape of the spectrum best, hopefully the process can be automated soon. I averaged about 200000 spectra or 2-3 minutes of exposure. Depending on your system temperature, a shorter exposure time should be good enough to detect the signal. All these are done in python with the wrapper pyrtlsdr.

A good way to check if you have detected is do to a 24 hours drift scan like what you've mentioned. When the milky way passes through the beam, signal rises, when it exits, signal falls. I have attached the gif below, this is done with the clone rtlsdr, lots of spikes. You can try to examine the raw spectrum, the spectrum before baseline subtraction. There should only be a single DC spike at the center of the spectrum, and the rest could be RFI that you can try to identify. 

Another good website to verify your signal is here https://www.astro.uni-bonn.de/hisurvey/euhou/LABprofile/index.php, your peaks should be very similar to to simulated spectrum.

Zi Hao
 

[Victor Boesen]

3 hours ago, SteveBz said:

Does anyone have any idea what is causing them?

Could be many things unfortunately!! PC noise, power supply noise, LEDs, flourescent lamps, the list goes on. A quick fix can sometimes be adding some ferrite clamps around the coax- and USB cable. It's not guaranteed to work, but sometimes it does if the noise is coming from USB or the PC.

[SteveBz]

1 hour ago, ZiHao said:

Hello Steve,

When I was using the clone rtlsdr, I used to take a reference spectrum at another frequency, where there is no hydrogen line signals, and subtract that with the spectrum with hydrogen line signal. It should be clear that whether or not there's a signal, despite the clone dongle having a lot of noise spikes. With the genuine rtlsdr, I am trying to fit a polynomial curve to the source spectrum to avoid adding extra noise when subtracting. The result I posted is manually done by eyeballing the curve that fits the overall shape of the spectrum best, hopefully the process can be automated soon. I averaged about 200000 spectra or 2-3 minutes of exposure. Depending on your system temperature, a shorter exposure time should be good enough to detect the signal. All these are done in python with the wrapper pyrtlsdr.

A good way to check if you have detected is do to a 24 hours drift scan like what you've mentioned. When the milky way passes through the beam, signal rises, when it exits, signal falls. I have attached the gif below, this is done with the clone rtlsdr, lots of spikes. You can try to examine the raw spectrum, the spectrum before baseline subtraction. There should only be a single DC spike at the center of the spectrum, and the rest could be RFI that you can try to identify. 

Another good website to verify your signal is here https://www.astro.uni-bonn.de/hisurvey/euhou/LABprofile/index.php, your peaks should be very similar to to simulated spectrum.

Zi Hao
 

So I have almost the same spikes as you.  How can you tell if the rtl-sdr is genuine or not?

Steve

[SteveBz]

28 minutes ago, Victor Boesen said:

Could be many things unfortunately!! PC noise, power supply noise, LEDs, flourescent lamps, the list goes on. A quick fix can sometimes be adding some ferrite clamps around the coax- and USB cable. It's not guaranteed to work, but sometimes it does if the noise is coming from USB or the PC.

So I did try shortening the coax from 25 m to 5 m, so maybe it improved.  It was before I found the missing pins

Maybe I could buy a handfull of ferrite clips and put them everywhere.

Steve

[SteveBz]

But Victor, the 24 hour scan is nearly done and it worked quite well.  If I ignore the spikes there are some nice patterns.  Can I ask, are you both using the NooElec H1 Premium LNA and the RTL-SDR v3 Blog SDR?  If I add another LNA, it will increase the background noise together with the signal, but not the SDR spikes, I guess.

I could also increase the n= number from 10,000 to even higher.  The problem is that it means fewer increments, so the time is now 8.5 minutes per sample of 10,000 which is 144 steps of 2.5 degrees.  As the beam-width is about 10-12 degrees, the longer the sample time, the less accurate the resolution.  2.5 degrees is 20-15% of the beam width, so I don't really want to make it less precise.  If I had a longer sample time I think I'd need tracking!

Not sure what to do next.

Kind regards

Steve.

Edited March 18, 2022 by SteveBz

[ZiHao]

23 minutes ago, SteveBz said:

So I have almost the same spikes as you.  How can you tell if the rtl-sdr is genuine or not?

Steve

https://www.rtl-sdr.com/buy-rtl-sdr-dvb-t-dongles/ Scroll down and you will see the picture. But if you were able to use the bias tee, most likely it is genuine. Then the spikes should be interference.

[Victor Boesen]

1 hour ago, SteveBz said:

Can I ask, are you both using the NooElec H1 Premium LNA and the RTL-SDR v3 Blog SDR?  If I add another LNA, it will increase the background noise together with the signal, but not the SDR spikes, I guess.

Yes, I am using both the Nooelec H1 SAWbird LNA and RTL-SDR V3. Adding another LNA will increase system noise floor as you say, and I would advice against it. After all, it's not certain the spikes are due to the SDR. Often spikes from the SDR only occur at harmonic frequencies of the local oscillator in the SDR for example.

I would stick to n=10k for the reasons your mentioning yourself. Running observations locally on a device (without RTL-TCP), however, will be a lot faster than the 8.5 minutes (closer to 2 minutes from my testing).

I would try to get a couple ferrite clamps since they can be found quite cheap and would be an easy fix. If that doesn't help, then there are only a couple options left to explore: Either try to locate the source of the noise spikes or move the antenna to a place less affected by this noise.

Victor

[SteveBz]

2 minutes ago, Victor Boesen said:

Running observations locally on a device (without RTL-TCP), however, will be a lot faster than the 8.5 minutes (closer to 2 minutes from my testing).

Interesting.  I may try that.  The problem is that I don't have much space left on my RPi.

[SteveBz]

Trying to upload the GIF, but it's crashing.  It's 213 MB, so I guess there's a size problem.  I might try again later.

[SteveBz]

21 minutes ago, Victor Boesen said:

Running observations locally on a device (without RTL-TCP), however, will be a lot faster than the 8.5 minutes (closer to 2 minutes from my testing).

So if that means 2 minutes exposure and 6 minutes download, it means there is less 'star-trailing'

[SteveBz]

40 minutes ago, Victor Boesen said:

move the antenna to a place less affected by this noise.

Also a good idea, as maybe the noise is comming from all the electronics in the observatory!!

[Victor Boesen]

2 hours ago, SteveBz said:

So if that means 2 minutes exposure and 6 minutes download, it means there is less 'star-trailing'

For RTL-TCP each sample is streamed to the client computer immediately, so the entire observation is the same length as it takes from start to plot of data. So trailing can still be an issue The duration of observations through RTL-TCP will be limited to your wifi transfer speeds. That is why USB will be faster since it doesn't need to remain a connection between two TCP sockets.

[SteveBz]

Here is a shortened and compressed gif of a the period.  It's a bit noisy, but the central region around 1.4 is not too bad.

I've ordered a box of ferrite clips to put on everything in sight!

Steve.

Edited March 19, 2022 by SteveBz

[SteveBz]

OMG. I just unplugged all the electronics in the observatory! Look how clean it is! No Spikes at all!

Edited March 20, 2022 by SteveBz

[Victor Boesen]

1 hour ago, SteveBz said:

OMG. I just unplugged all the electronics in the observatory! Look how clean it is! No Spikes at all!

Excellent!!! That's a great observation 

[SteveBz]

And here is 24 hour's worth.

I had to resize the photos to make it work with:

mogrify -resize 960x *.png

Which is available from ImageMagick. Ie
 

sudo apt-get install -y imagemagick

Now I need to start my campaign against the other sources of noise while waiting for the stepper motor driver to arrive.

Kind regards

Steve.

[Victor Boesen]

That's excellent Steve!!! May I ask what parameters you used? The spectrum is very good looking, far better than what I've managed!

Victor

[SteveBz]

1 hour ago, Victor Boesen said:

That's excellent Steve!!! May I ask what parameters you used? The spectrum is very good looking, far better than what I've managed!

Victor

Hi Victor,

Thanks for your kind words.  Here is the command line:

python3 H-line.py -s 2400000 -r 11 -n 10000  -e 192.168.1.116 -m 75 -i 2.50 -d -c

And here is the config file:

{
    "latitude": <x>,
    "longitude": <y>,
    "azimuth": 0,
    "altitude": 90,
    "low_y": -0.2,
    "high_y": 1.3
}

I've just got 20 of those magnetic collars you recommended in the post this morning.  I attached them all to everything in sight, but sadly it hasn't helped.  I think I need to step through each device one at a time to see where the noise is soming from.

Kind regards,

Steve

Edited March 22, 2022 by SteveBz

[SteveBz]

I should add that I also split out the 'gif' piece of your code into a spearate program so that I could run 'mogrify' first and then 'gif' it. Like this:

import imageio 
import os

# Main method
# def main(args):

# Generates and saves a GIF of 24H observations
def generate_GIF():
    arr_txt = [x for x in os.listdir('Spectrums/clean') if x.endswith("png")]
    dec=51.1
    ra='all'

    ras=[]
    for fileName in arr_txt:
        sub1=fileName.find('ra=')+3
        sub2=fileName.find('dec=')-1
        raFlt=fileName[sub1:sub2]
        ras.append(float(raFlt))
       
    ras.sort()
        
    print('Generating GIF from observations... This may take a while')
    path = f'Spectrums/clean/ra={ra[0]},dec={dec}.gif'
    images = [imageio.imread(f'Spectrums/clean/ra={str(raFlt)},dec={dec}.png') for raFlt in ras]
    imageio.mimsave(path, images)

if __name__ == "__main__":
    generate_GIF()

NB, I put all the files I want to 'gif' in a separate directory called 'clean'.

Steve.

[Victor Boesen]

Thanks!! I would watch out increasing the median smoothing too high as that could hide some of the faint/weak details of the hydrogen line such as individual peaks and etc. But other than that, I'd say it looks great

[SteveBz]

1 hour ago, Victor Boesen said:

Thanks!! I would watch out increasing the median smoothing too high as that could hide some of the faint/weak details of the hydrogen line such as individual peaks and etc. But other than that, I'd say it looks great

Yes.  I tried it up to 200 and I felt 75 was an OK value.

[SteveBz]

Hi Guys,

So I still have a noise problem, even with all the electronics powered down.  Most of my scans look like this:

But then occasionally, I get one of these:

And everything seems worthwhile 😀

Maybe it's the central heating system or the fridge.  Maybe I can move the dish a bit further away (or just turn the fridge and central heating off - not sure what the family will think).

Kind regards,

Steve.

[Victor Boesen]

53 minutes ago, SteveBz said:

So I still have a noise problem, even with all the electronics powered down

It's difficult to conclude what should be to blame for this noise. One option is to try and locate the noise source with something like a yagi antenna, which you can point in different directions and look for noise sources.

[SteveBz]

And here is the maximum intensity in dB returned for each RA vertically overhead. Sorry it's in Excel!!  The peak at about ~75 deg is much wider than the peak ~320 deg, which corresponds to the breadth of the curve on the diagram from the H1 survey.  Obviously when I get the DEC axis working I'll need to get it into Matplotlib to get the 2D map, and the peak frequency could be false colour to show the doppler effect, like red <1.420 GHz (travelling away from us) and blue >1.420 GHz (travelling towards us).  I'd like to show the lobes too where there is more than 1 peak.  Not sure how to do this.

Sadly, I think the stepper motor I have to drive the dish may not be working, as so far it has resisted all my efforts.  I'll try for a bit longer, and then I may need to find a replacement.  Not sure how, I may reach out here!!

Hope you're having a good weekend.

Kind regards,

Steve.