How RADARs use CFAR to detect targets

[@SinanAkkoyun]

This is peak education: Covering every single detail mathematically and intuitively WHILST providing an interactive notebook to play with

[@MarshallBrunerRF]

More notebooks to come! The next video has a really exciting one. Stay tuned!

[@HoffmansHoopties]

A full 1/3 ratio of likes to views is absolutely bananas and a flashing red sign that you’re onto something! Great vid. Love the 3Blue1Brown vibes. 🎉

[@MarshallBrunerRF]

Thanks so much for your support! It's been really nice to see that people have liked it

[@jannegrey]

Even now it's 10%

[@connerray3872]

I was curious if he ended up using 3Blue1Brown's library for this. It would be a nice trend if more people used it or similar to put out these videos as it is quite excellent as a format.

[@MarshallBrunerRF]

@@connerray3872 yeah I’m using manim! Awesome library

[@MrSomethingdark]

nah don't be like 3Blue1Brown. Be what you like. Do the explaining you think we should get. Btw I love it way better than 3Blue1Brown

[@find2hard]

I predict with high confidence that this channel will produce more quality content.

[@MarshallBrunerRF]

Thanks so much for watching! Another video in the works now. I'm very excited for this next one!

[@user-tm1kb3fq4m]

@@MarshallBrunerRF Will you be making vids on EP or EW techniques?

[@MarshallBrunerRF]

I will not be making content on EW, but I will cover a lot in the radar space. Thanks for watching!

[@AdrianBoyko]

I’m only tangentially interested in radar but I’m glad I watched this. I can think of a lot of non-radar situations where this will be useful.

[@joet4348]

It is a really cool idea. Could I ask what other applications you had in mind?

[@fatant]

@@joet4348finding energy peaks in compton scattering

[@c4llv07e]

@@joet4348 Edge detection, for example. Don't know if it'll work, but I think it could.

[@MarshallBrunerRF]

Yeah there's many applications for this!

[@HipsterBaby747]

It’s a great day when a new channel like this pops up

[@MarshallBrunerRF]

So glad you enjoyed :)

[@MrSomethingdark]

My RIO skills in the F-14 in DCS have improved after this video as well as my programming knowledge! Thanks. Will maybe try this in C

[@MarshallBrunerRF]

Thanks! Definitely try it in C! Best way to learn

[@electrowizard2000]

When I worked on a weather radar, we sampled receiver noise and tuned the detection parameters to a specified false alarm rate, Monte Carlo style. Was a fun project. Would have liked a bit more math in the video about the actual false alarm rate that the algorithm produces, but still great work thanks.

[@MarshallBrunerRF]

Yeah I struggled with whether or not to include more of the stats portion of the false alarm rate but ultimately decided not to in the interest of generality. I did include some books and resources in the description that talk about this, though. Thanks for watching!

[@itsmemakz]

This was such a fun and enlightening video, You got me so excited that i coded it up myself and used it to detect peaks in local difference measurements in the comparison of protein structures. Great explanation, geat video.

[@MarshallBrunerRF]

Hey! That's the goal! So glad you went and coded it up on your own

[@reggindog3436]

I am really enjoying this series, I stumbled on it by accident. One thing that you may want to cover more here, and you kind of joked about, was if you knew the distance, why would you need a radar? In a lot of cases you sort of know the distance, so that should let you chuck out a lot of unrealistic results. An awful lot depends on the use, but say in a machine, the distance to be measured is never going to be outside of some specific range. For example using the radar to replace an lvdt and getting rid of the mechanical component that in theory is more likely to fail. In say a car's adaptive cruse, again you have a range. You do not need to slow the car down until it is within a certain range, and you worry about that range to when the cars contact each other and try to avoid that. The other thing is, in one video back, you described how to fm modulate a pll. That is one way to do it, but you can also apply the ramp by adding it to the control voltage of the VCO,. This takes less electronics, and the output does not have steps in it.

[@MarshallBrunerRF]

Hi thanks for watching and for your comment! Yes there are so many different applications for radar with varying requirements. You can definitely purpose-build a system if you don't care about certain ranges, velocities, etc. As for the PLL: yes there are other ways to do it including the way you mentioned and direction conversion, but I just showed one. My goal is to get people a basic understanding so they can go off and explore for themselves. Thanks again!

[@lorenzobertolino1527]

What a great video and series! Fantastic animations, excellent explanation and interesting subject Immediate subscription 🤩

[@MarshallBrunerRF]

Awesome! So glad you enjoyed it

[@NaifAlqahtani]

Really good video with an undermining title. Really glad I still clicked to watch regardless of the title

[@MarshallBrunerRF]

What would you have called it? Thanks for watching!

[@NaifAlqahtani]

@@MarshallBrunerRF try not to include technical words. I have never heard of CFAR before so I wouldn’t care about how to use it, let alone about its application in RADAR technology. A more general title would be applicable. Something like “How to differentiate between a spike and noise in data” or something along those lines. I dont like my title either tbh but it points towards the general vicinity of what I mean. However poor my title is, I would still be REALLY intrigued in finding out about it. The thumbnail conveys this PERFECTLY which is why I ended up deciding to click the video to watch. Good luck man!

[@NaifAlqahtani]

@@MarshallBrunerRF see @AdrianBoyko’s comment as well where he mentioned that he only slightly marginally cares about RADARS but is still glad on clicking to watch. His comment has 21 likes so I assume 21 others also agreed with him. Hope this helps

[@MarshallBrunerRF]

That totally makes sense. Thanks for your input :)

[@diegobrsrkk]

Hey! Cool videos. I am a Satellite Antenna Engineer (junior) and I love this niche videos.

[@MarshallBrunerRF]

So glad you liked it!

[@exXwULrFni]

Thank you very much. It's much more better than mmwave CFAR explanation

[@Horatio_101]

Great video. Right now, I am pretty spent after work, and most videos like that then go to "watch later", but you managed to keep me hooked nevertheless!

[@MarshallBrunerRF]

Haha yeah sometimes the technical videos are too much at night :). Glad you enjoyed it

[@liamaenlle]

This video was excellent, way to go man! Super polished and well constructed. The stats at the end of the video are so cool!

[@MarshallBrunerRF]

thanks! Glad someone liked the stats :)

[@VijayKumar-vk1en]

Excellent Explanation, it took me many days to understand this. But your animations makes us understand this concept in minutes. Great animation.

[@MarshallBrunerRF]

Great! So glad it helped!

[@chalybesmith]

Amazing video! More signal processing videos please!

[@MarshallBrunerRF]

More to come for sure! One was posted today :)

[@rfdspguru2]

Kudos to you and Sabrina for another great video. I'm casting my vote... I'd love to see Manim on a 4D datacube (over steering angle)

[@MarshallBrunerRF]

Thanks! The video I'm working on now is about range-Doppler measurements and I plan to talk about at least the 3d radar cube (range, Doppler, channel).

[@EdoKarachannel]

I want to echo what others have said here and say this is an extremely good explanation and a great video! I do a lot of signal processing right now and can see a LOT of potential to apply this kind of algorithm to other domains...

[@MarshallBrunerRF]

Yeah it's used in many areas and applied in different ways. Glad you found it useful!

[@electric_sand]

Hi! Thanks for the videos. Video idea - From your channel bio, I note that your intention is to introduce the field to people. To achieve that, I suggest you show the layout of the land i.e the prerequisites, application areas, recommended books etc Something to get people started. Best regards.

[@MarshallBrunerRF]

Yes! I would love to make a video like this. Good to know you would find that useful. It's added to my ideas list

[@ChrisAthanas]

Very nice breakdown of a complicated topic

[@MarshallBrunerRF]

Thanks!

[@PauloKuga7]

one of best videos i ever seen

[@MarshallBrunerRF]

Woah thanks!

[@Sniffy.Gumbles]

New channel is batting 1000 for quality videos, instant subscribe, looking forward to your future stuff.

[@MarshallBrunerRF]

Awesome! Thanks for watching!

[@exceedinglycurioable]

This was great I hope you maoke more videos on RF algorithms.

[@MarshallBrunerRF]

I certainly will! Working on one now....

[@gameshot911]

Really great animation transitions!

[@MarshallBrunerRF]

Thanks! Much thanks to the manim community

[@AngryMarkFPV]

Thank you for this video sir!

[@uumuu]

I can't exactly explain it but this feels like a special case of wavelet convolution

[@MarshallBrunerRF]

There's lots of variations and similar algorithms, I wouldn't at all be surprised if it resembled the wavelet transform. Thanks for watching!

[@rahulkushwaha9500]

the animations are sooo good, thanks

[@MarshallBrunerRF]

Thanks! Manim is such a cool library

[@shadowdragon2484]

I love how many resources you provide -- really great!

[@MarshallBrunerRF]

Great! Hopefully you get some use out of it. I'll keep adding them

[@isodoublet]

Not necessarily advocating for a constant threshold here, but it's possible to compute a threshold that's optimal for your data by using the Otsu thresholding technique from image processing (since it operates on the intensity histogram only the dimensionality doesn't matter). You essentially pick the threshold value that minimizes the intra-class variance, and that will be the optimal choice for that particular signal.

[@MarshallBrunerRF]

Yeah there's tons of options, this is just a cool one I've worked with. Thanks for the comment!

[@duncanmcallister7932]

This is awesome!!

[@MarshallBrunerRF]

Thanks so much!

[@joet4348]

Dope video, thank you. Subscribed

[@a11y93]

Great video with some truely impressive, imformative graphics!

[@MarshallBrunerRF]

Thanks for watching! Yes there's previous videos but I definitely think I could dive into this topic of where the data comes from more deeply. Thanks for the critique

[@MorRobots]

Now you understand why modern radars all have FPGA front ends doing DSP, because that's essentially just a 1D convolution kernel.

[@MarshallBrunerRF]

Yup, it really looks like a CNN when you do it on 2D data

[@felix6142]

the details from this video should be added to the cfar wiki page, its so barebone :D

[@MarshallBrunerRF]

Haha yeah it is! Thanks so much for your comment!

[@JoeCoup1]

your manim is really good!

[@MarshallBrunerRF]

hey! Thanks! I liked your video on regression too

[@rafa_br34]

Very interesting indeed.

[@TheBauwssss]

Holy fuck this is AWESOME.

[@MarshallBrunerRF]

Hahaha thanks!

[@TannerMageeYT]

Another banger

[@MarshallBrunerRF]

Hey! Thanks!

[@0xc0ffee_]

I was looking for some ideas on how to find peaks and filter out noise when working with audio buffers coming from songs. I was doing an algorithm for peak detection

[@takyk_bilim]

Amazing 😍

[@MarshallBrunerRF]

Thanks for watching!

[@ulysisxtr]

Phase array

[@MarshallBrunerRF]

I will definitely make some videos on this topic :)

[@EigenA]

Beautiful

[@reeeeeeee2143]

is this not similar to convolving a wavelet filter across the signal?

[@MarshallBrunerRF]

@@reeeeeeee2143 basically yeah!

[@adissentingopinion848]

I feel like the gap could be replaced with multiplying by a negative gaussian curve +1, but that sounds like a variant that ought to have been implemented somewhere.

[@MarshallBrunerRF]

Yes, there's tons of variations of this, of which I only covered a few. Thanks for watching and commenting!

[@aerialbugsmasher]

PLEASE work on taming your upward inflection, also known as uptalk or high-rising terminal. That would make the video narration much better.

[@MarshallBrunerRF]

Thanks for the suggestion!

[@EigenA]

For the N in your denominator, is that N w.r.t your entire signal or just the gap, cut, and ref elements?

[@MarshallBrunerRF]

Just the ref elements. You're just getting the mean of the ref cells. Thanks for watching!

[@EigenA]

@ thanks for making!

[@MarshallBrunerRF]

Of course! It's fun interacting with everyone

[@danamassie]

Can you talk about Projection into noise sub space versus signal sub space.

[@MarshallBrunerRF]

That would be an interesting video! Thanks for the suggestion!

[@_ifly]

please do some tutorial like this for FMCW radars such as Ti's radars. thanks for the video

[@MarshallBrunerRF]

In the video coming out next week on measuring velocity with FMCW waveforms, I specifically use a TI radar to show how it works in a real system. Thanks for watching!

[@_ifly]

@@MarshallBrunerRF Thank you so much; it truly means a lot. I am currently working with the TI IWR1443 radar and need to perform signal processing. Since I am new to this field and there are no readily available tutorials, I would greatly appreciate your help. Thank you once again!

[@MarshallBrunerRF]

@@_ifly hope you like the new video! It's out which uses the AWR1642

[@toastrecon]

Trying to think about how you implement or add doppler shift to this? Probably because I've watched too many DCS World videos, but I know that military aircraft use that to find "fast" targets among static or slow clutter.

[@MarshallBrunerRF]

Next video is on the FMCW range-Doppler spectrum. This algorithm can and has been been extended to 2D. Thanks for watching!

[@toastrecon]

@@MarshallBrunerRF Very cool! Thanks!

[@EigenA]

Keep it up

[@karthikraj6347]

Passive Radar!

[@MarshallBrunerRF]

That's a super interesting concept I would love to dive into more!

[@nicolaiverheul]

More signal analysis! How to make non stationary analysis with bayesian time series

[@MarshallBrunerRF]

More signal processing to come!

[@ozne_2358]

I imagine that a small neural network, trained on real data, sliding along the samples has been done already, right ?

[@smoshGaming]

Radenso was doing this a while ago to detect false radar emissions on their radar detectors. They quickly got government contracts and abandoned this for the civilians.

[@MarshallBrunerRF]

Yes, there's lots of research in this area! Very exciting stuff

[@rezah336]

just sliding window mean, find outliers, remove outliers and repeat until no outliers, this would detect all these 3

[@MarshallBrunerRF]

Yeah, there's so many detection options, but this is a cool one that is relatively fast to implement. The one you described seems to require looping over the data multiple times which would be expensive in a real-time signal processing environment. Thanks for watching!

[@sbreheny]

I don't think you explained why it is desirable to make the false alarm rate constant as opposed to, say, minimal. My understanding is that it makes interpretation easier. If you know the false alarm rate, you know that seeing detections more often than that means that there probably is a real target, whereas if the false alarm rate were lower on average but variable, you'd need more time to tell the difference between likely targets and false alarms.

[@genericalias5756]

isn't this very similar to just applying a laplacian of a gaussian filter?

[@MarshallBrunerRF]

From my brief google search, yes it looks similar. I'd be curious to see how that looks on the same data. That'd be a cool exercise if you want to try it. Thanks for watching!

[@JonathanFraser-i7h]

That's still just subtracting a low-pass, it's just a specially shaped lowpass.

[@MarshallBrunerRF]

Basically...but with some different design knobs

[@ir0nsight280]

My early warning radar confused a flock of birds with a nuclear first strike attack and launched counter missiles :( help

[@MarshallBrunerRF]

unfortunate

[@Livenewme]

Yeah this is cool and all, but why is there no banana for scale

[@MarshallBrunerRF]

My sincere apologies. I think I will have to delete this video and re-upload with a banana for scale.

[@arbitrandomuser]

[@sbreheny]

I think you also never show HOW this algorithm results in a constant false alarm rate across the range. I think you need more information and assumptions about the probability distribution of the noise and clutter.

[@MarshallBrunerRF]

I included some resources and books in the description that cover this. Decided it was too much to include in the intro video. Thanks for watching!

[@distrologic2925]

Once again: all you need is context

[@MarshallBrunerRF]

That's why I'm providing resources and the python notebook to play around with the concepts yourself. Hopefully this is enough, but let me know if there's anything else that would improve yours and others' understanding. Thanks for watching!

[@factoryofgaming5229]

My first thought: How can we apply this to financial markets?

[@MarshallBrunerRF]

I'm sure it could!

[@whateverrandomnumber]

6:26 Nivel de emprego, geração de renda? É uma indústria, amigo. Só gera aumento de renda pros donos dos conglomerados.

[@sergehog]

Looks like over-engineered. I don't see why it shall be better than "dynamic thresholding", you explained in the beginning. Did anyone ever compared them? Same performance could be achieved with low-pass filter, which has somewhat weird kernel.

[@Aubstract]

It honestly doesn’t seem that complicated to me. And it can be easily parallelized at a hardware level on a device like an FPGA, which are very commonly used in signal processing. So it would be fast too.

[@MarshallBrunerRF]

Exactly! The implementation must always be considered too. These systems often are running as quick as possible on low-power hardware rather than in some research environment on a PC using python like I showed here.

[@robertthallium6883]

I am so beyond dumb. Really wish I watched Star Trek at 7 instead of 31. Fuck this is insanely onion layered in background knowledge required just to find the first step.

[@MarshallBrunerRF]

Maybe playing with the python notebook in the description would help your understanding. Thanks for watching!

[@jhacklack]

"noise from lots of sources like...lots of others" Please edit your script better.

[@MarshallBrunerRF]

thanks for watching