Lightnews — Scholar-powered news

hforsten @hforsten.com · 16m

Python, C++, and CUDA. github.com/Ttl/torchbp

GitHub - Ttl/torchbp: Fast C++ Pytorch extension for differentiable synthetic aperture radar image formation and autofocus library on CPU and GPU

Fast C++ Pytorch extension for differentiable synthetic aperture radar image formation and autofocus library on CPU and GPU - Ttl/torchbp

github.com

hforsten @hforsten.com · 1d

New blog post: Synthetic aperture radar autofocus and calibration.

Synthetic aperture radar autofocus and calibration

3D trajectory position error estimation autofocus, antenna pattern normalization, and polarimetric calibration for drone mounted SAR radar.

hforsten.com

4 11

hforsten @hforsten.com · 3d

SAR image is processed offline. It's just a hobby project.

2

hforsten @hforsten.com · 4d

Synthetic aperture radar vs Gopro.

3 3 26

hforsten @hforsten.com · Jul 13

Wrote code for SAR image antenna pattern normalization and polarimetric calibration (channel imbalance and crosstalk). Image looks very nice now.

Same image after antenna pattern normalization and polarimetric calibration.

3 3 26

hforsten @hforsten.com · Jun 12

It's 10 seconds of data (10,000 radar pulses) and it takes about a minute to process.

2

hforsten @hforsten.com · Jun 11

I implemented a new SAR autofocus algorithm and now the image generation is both better and faster. It's about as well focused as it can be now.

Polarimetric synthetic aperture radar image

2 13

hforsten @hforsten.com · Jun 5

Fixed some bugs in my SAR processing code and now fast factorized backprojection generates a visually identical picture to the normal backprojection, but over 10 times faster. That's huge because normal backprojection was already extremely quick and there's still room for optimization with FFBP.

15

hforsten @hforsten.com · Jun 1

I also implemented generalized phase gradient autofocus earlier this week. I haven't seen any open source implementations of it before. It works quite well with this kind of short range SAR image, but my earlier minimum entropy optimization method still gives better results.

7

hforsten @hforsten.com · Jun 1

Found motivation to write some SAR processing code. Fast factorized backprojection generates several small polar format images and then interpolates them to one big image. It's faster, but has slightly worse image quality due to all the interpolations.

Global backprojection vs fast factorized backprojection image quality. The biggest difference is larger sidelobes on the brightest target in the FFBP image.

1 2 18

hforsten @hforsten.com · May 30

Just some random ones I had lying around. The bigger one is labeled: "XRRH6*10*3"

1

hforsten @hforsten.com · May 29

That's the measured S21 from input to the isolated port when the other ports are terminated. Reflection of the termination is not de-embedded, but it's better than -25 dB over the whole frequency range.

2

hforsten @hforsten.com · May 29

I made this new dual directional bridge coupler using 1 mm diameter coaxial cable. It's a little bit better than the previous one I made using 2 mm coaxial cable and much more annoying to solder.

Thru, coupled, and isolated port S21 with other ports terminated.

2 10

hforsten @hforsten.com · May 28

It works really well. Loss is low and matching is excellent.

5

hforsten @hforsten.com · May 28

Bought this thru with 3.5mm RF connectors. It's air filled with just a thin see-thru membrane holding the center pin in place.

2 4

hforsten @hforsten.com · May 21

Some new RF connector test PCBs. Nameless low-cost solderless 3.5 mm connector on the left and CONSMA003.062 clone on the right. Both work okay.

Two test PCBs with 3.5 mm and SMA connectors

3.5 mm and SMA connector S-parameters with 11 mm line

3.5 mm and SMA connector time domain response with 11 mm long line

1 7

hforsten @hforsten.com · May 15

It would require hole and threads in the case for the connector making the case more expensive. Connector would also need to be soldered to the PCB while attached to the case, not very handy for a prototype.

hforsten @hforsten.com · May 6

It's this one: www.aliexpress.com/item/1005006.... Calibration coefficients are not included.

www.aliexpress.com

1 1

hforsten @hforsten.com · May 6

Bought this VNA calibration kit with 3.5 mm connectors. Looks good considering the 60€ price tag.

2 1 4

hforsten @hforsten.com · Apr 15

It's a very nice looking chip, but too expensive. Sure if someone else pays.

1

hforsten @hforsten.com · Apr 15

New blog post: "Designing a low-cost high-performance 10 MHz - 15 GHz vector network analyzer".

hforsten.com/designing-a-...

Designing a low-cost high-performance 10 MHz - 15 GHz vector network analyzer

Designing a cheap two-port vector network analyzer with good measurement accuracy.

hforsten.com

3 3 13

hforsten @hforsten.com · Apr 10

Just found an annoying flaw with this cheap bias-T. The connector is threaded in the case and if the cable is tightened with a torque wrench, it can unscrew also the SMA connector when removing the cable.

hforsten @hforsten.com · Jan 23

I bought this cheap Bias-T from China. It claims to function from 10 MHz to 10 GHz. Looks very nice from both outside and inside. Enclosure is machined from aluminium and PCB is not FR4. Measured with VNA it seems to work fine up to 6 GHz, but there's a small 1 dB dip in S21 around 1.1 GHz.

1 1 4

hforsten @hforsten.com · Apr 5

That could work in some situations, but I don't think it will work here. The connection between SMA connector and the case is not tightened so it wouldn't be compressed. Mating surface also isn't a full circle, just the upper part of the connector.

1 1

hforsten @hforsten.com · Apr 5

I thought this might be an issue during design. The proper fix would be to use a bulkhead SMA connector but that complicates the mechanical design and I didn't think it was worth the trouble.

2 4

hforsten @hforsten.com · Apr 5

There's a tiny gap between the case and SMA connector that radiates a little if not sealed. Putting some solder wick in the gap seals it nicely but there has to be some more elegant method.

2 11