Видео

One way is to reduce the mesh size until simulation results do not change much with reducing mesh step. For example, S11 resonance 6.5 GHz, 6.1, 6.05, 6.01, 6.015. There are many recommendations on classical rectangular FDTD mesh size. But usually simulation time becomes too long earlier. Currently, for larger models I try to share meshing lines between the elements. This puts some constraints on the geometry though

Thank you, with all letdowns I had recently your comment means a lot and brings more light into my life!

@@TechSponsorTV happy to hear that. What letdowns if i can ask?

more than a year trying different things with very little financial gains and some personal stuff adds up to this. I think that I have enough focus, just feels that I am moving very slow..

@@TechSponsorTV don't i know that feeling :D you would think with RF knowledge it would be a breeze to make money, but the times are hard

Thats unexpected and Thank you!

I like something about your music, but I don't know what exactly!

@@TechSponsorTV this might be one of my favorite responses ever about my tunes. That means to me I’m being authentic in some way. Thanks so much :)!

You're smart! I have used an AGC chip for amplification. That's why recording looks better. Automatic gain control significantly widens dynamic range of STM32F103's 12-bit ADC. Spectrogram is also normalized for visualization purposes. Engineers usually avoid AGC for such applications, because it's no longer possible to use amplitude thresholding algorithms. But here I use time threshold, so it still works pretty good and range is extended too. Georgy

to what board you are connecting it to? I have used stm32 board and connected vin to 3.3v, gnd to gnd, scl to scl, sda to sda, add to gnd. So, there are two wires for power, two wires for i2c and one wire for address pin to make it 0x28. I haven't tested serial connection, but i2c definitely works. Haven't tried arduino with its 5v either. This board doesn't have mosfet bidirectional level conversion circuit, and sda/scl are pulled up to a 3.3v from LDO output. Not sure if it works with 5v arduino reliable with such i2c input circuit.

@@TechSponsorTV welp I tried both on UNO and nano 33 iot boards. Connect SCL to analog 5 Connect SDA to analog 4 Connect VDD to 3.3-5V DC Connect GROUND to common ground Didn't work, tried different cables and computers. I did try adding the add wire to positive railjack on my bread board and I2C 0x60 was found. I am switching to qt py esp32 s3 with BNO055 + MP BFF combo since the project I am working on is small. Gonna return my stuff soon.

Probably one of the best performance at 24 GHz are old UMRR series, for example UMRR-11 by smartmicro. But it's an MFSK. What are you going to measure? Generally,I would recommend any radar front end capable to control VCO in wide enough bandwidth B, because resolution R=C/(2B) can be rediculously low. With SLOW FMCW waveforms it is problematic to have good resolution and stay in allowed bands. FAST FMCW waveforms are used in modern 70GHz automotive radars, maybe you can find some model with documentation.. I am not sure, need some more info on your task. For example, with 250 MHz bandwidth resolution is around 0.6 meters. In real 10GHz FMCW traffic counter allowable bandwidth was lower, and as I recall there was only 2 fft points per lane, meaning resolution around 1.5 meters. Do not know any good affordable modules lately.

Check pins in cube, maybe something is shorted out on your board due to pin assignment. Also, check if my sketch uses same MCU. I may have used STM32H7B0 and not H750. Sometimes binary from similar series work, but clock registers may be incompatible and cause overheat as a result of wrong PLL settings

Thank you, to everyone: Source code for this video on my GitHub

Yes, the most popular method is to use multiple Rx antennas with phase monopulse method. E.g., two Rx antennas spaced one-wavelength apart can be used for unambiguous angle measurements in -30 to +30 degrees range or maxAng=2*Arcsin(lambda/(2*d))=60 when d=lambda. This will provide angle PHI to the target. And the distance R is measured using FMCW, FSK, Multiple-frequency FSK (MFSK). Then we can obtain target 2D coordinates as x=R*Sin(PHI); y=R*Cos(PHI). Here is how it works for FMCW: Single Tx antenna emits microwave signal with frequency jumping between F1 and F2 at 100kHz, which also limits unambiguous range measurement to maxDist=C/(2*(F2-F1)). Signal bumps into some target travelling at speed V and reflects with some Doppler shift. This reflected signal will be received by both antennas Rx1 and Rx2. If target is on the left, then Rx1 receives signal a little earlier than Rx2, and exact target angular position can be calculated from phase difference between receiving mixer outputs of antennas Rx1 and Rx2. Note that mixer outputs will provide crazy meandered FSK signal at 100kHz, and before working with it one more thing should be done. Older radars used sample-and-hold circuit that switched at 100kHz (or something like that) to separate received signal to two non-meandered signals. Newer radars simply use ADC working at 100kHz and synchronized with FSK transmitter switching time. ADC sampling frequency can have an additional offset dt=0.5*1/100kHz in time to ensure that we do not sample the part during frequency transition. If each receiving antenna has a single-ended mixer, then down-converted signals Rx1 and Rx2 will be transformed to Rx1_F1, Rx1_F2 and Rx2_F1, Rx2_F2. If there are quadrature (IQ) mixers, then we will have Rx1_F1_I, Rx1_F1_Q, Rx1_F2_I, Rx1_F1_Q and Rx2_F1_I, Rx2_F1_Q, Rx2_F2_I, Rx2_F1_Q. Next step is to perform FFT. We can perform FFT on single antenna signals to find a distance. E.g., we can take FFT(Rx1_F1), find peak spectrum_Rx1_F1[N] with maximum amplitude. And find distance to the target from phase difference (phase(spectrum_Rx1_F1[N])-phase(spectrum_Rx1_F2[N])). Phase is found using ATAN2 function on bin [N] of complex spectrum. We can find the same distance using data from second antenna Rx2, perform averaging, etc. Next step is to find angle to the target: Now we compare phases of spectrum_Rx1_F1[N] and spectrum_Rx2_F1[N] (signal that was extracted for F1 frequency). The same angle can be found using spectrum_Rx1_F2[N] and spectrum_Rx2_F2[N]. Again, it may be beneficial to use all available data to find this angular value multiple ways and then perform averaging or something like that. If we have multiple targets with close speeds, then we will start observing various glitches: e.g., if FFT amplitude peaks spectrum_Rx1_F1[N1] spectrum_Rx1_F1[N2] are close, but targets angular positions or distances are very different, we will start to observe "averaged" ghost targets that are traveling somewhere in-between. That's why both FMCW and FSK are often replaced with Multiple-frequency FSK in many modern automotive radars. Here is another example that is easier to understand. An obscure method is to use two FSK radars placed at some distance apart. We can measure distance to the same target using two different radars: R1 and R2. If we draw two circles, one with radius R1 with center at (X1, Y1) and another with radius R2 with center at (X2, Y2), then these circles will intersect at target position. There may be multiple intersections, and detection performance can be somewhat optimized by increasing number of radars and their position. Because we also know radiation pattern of each radar, we can filter out circle intersections that occur behind the radar. About method from this video: it uses purely unmodulated radar with a single antenna. While very limited, it provides shockingly good results if targets have straight trajectories and constant speeds (or if there is a known model of target movement, that can be written as a formula to use for spectrogram pattern matching). Georgy M.

If we are going to measure relatively large objects like cars, most of the time there is no difference. But in some scenarios, geometrical features of objects can reflect vertical-polarized waves better than horizontal-polarized. E.g., some objects can act as parallel-plate waveguide, a giant "antenna". It can be a good thing if we want to see "behind" the target if signal was "guided" under the target (vehicle). Cons: we can have some strong unwanted reflections, signal modulated by vibrating/moving parts multiple times, etc. Very good but complex and expensive solution is to use a fully-polarimetric radar with reconfigurable beamwidth and active beam scanning. I.e., radar which is capable of transmitting horizontal, vertical or even circular polarized waveforms, and analyze how polarization changed after being reflected by the target. This answer probably not very useful, so here is some thoughts based on my practical experience: I have some favorite antenna configuration, currently it's an inline series-fed metamaterial antenna. If I need a wide-band security radar antenna with very wide beam in horizontal direction and very narrow beam in vertical direction, then I am limited to vertical polarization, or I need to design some other antenna (which I don't want to, or don't have time to, or tried and it's not as robust in terms of bandwidth, etc..). I can't simply rotate this antenna (need to meander a feeding line, which will affect bandwidth, etc.). Thus, I will look into polarization only if there are some very important advantages. During the years only two real-life applications regarding polarization catched my attention: first one is very noticeable difference in reflected signal amplitude from power line cables (vertical vs horizontal). And the second one is polarization change during each reflection of circularly polarized wave (e.g., LHCP becomes RHCP after reflection and you can deduce if signal was reflected odd/even number of times by using two receiving antennas: RHCP and LHCP to compare amplitude between them). And the third thing is related to beam-scanning and not polarization: if you have a giant rotating fan blades or some crazy vibrating surface it will modulate and reflect radar signal back, which will be seen as a huge interference on a FFT spectrogram. The only way to fight this is to prevent radar beam to illuminate problematic areas: e.g., if we have 3-degree beam rapidly scanning 60-degree area, it will illuminate problematic area only limited amount of time. If we have a passive 60-degree beam, it will be affected by interference all the time regardless of polarization. That's probably all I know about this. If you need more ideas - you may check papers on multipath for automotive radar and polarimetric radars for power line measurement, or even patents on these and similar topics. Georgy M.

I am glad that you've liked it!

People like your music!

No special reason. This video was recorded without me speaking and then uploaded. I guess it's a history already unless I record another video. I can ever sign a song maybe. Regarding FFT the most challenging part is making fast phase and amplitude measurements. CMSIS amplitude function loses dynamic range by right shifting and ATAN2 is very slow. I have solved these problems long time ago and will release a better tutorial soon but with a twist!

According to Udemy, students receive lifetime access regardless of what price they paid for it, or whether they used a coupon to enrol or not. In other words, you should be able to view the course and access any future updates anytime (for free). One month is coupon expiration date, but the course access is permanent. Hope this helps!

It's kind of boring until you have a prototype. I literally had goosebumps when realized that one of my first antennas worked. It becomes really boring when I over-optimize and tune the same design again and again; that's when it becomes a nightmare!

@@TechSponsorTV I can agree with you on that. However, if producing a design with that (improvement, regardless of how small), that 'brings in' that weak signal, and gets that QSO, error free data, and blocks out interference, YOU KNOW, it's all been worth the tears and frustrations!

Wow, great, fantastic. But Udemy won't accept a valid e_mail address. 😥 It is too good to be true, or is there a problem with Udemy account creation? 🤔

Is it possible that you already have an account on Udemy? The first thing I would try is to reset the password (even if I don't remember having an account).

@@TechSponsorTV What a surprise, you are correct! All fixed. 👍

Thanks for the offer! Have you thought about creating an Udemy course on this exact topic? I mean optimizing and stuff. So people can buy your information.

@@TechSponsorTV No, I work on fiverr as a freelancer I work there as a digital marketer

@@TechSponsorTV Let's work on how i can get your course promoted to get more sales on it

We are in the same boat here. I simply have no finances to pay you.

@@TechSponsorTV Did you know that after this job as being done you will have a lot of finance not that with a lot of sales

Great 👍

Some time later

@@TechSponsorTV we wait for you! 哈哈

I have received your message, hope you found someone to help with your project

@@TechSponsorTV not yet

it would be really helpful if you support

A first step to troubleshoot it is to take a small wire and connect ADC input to GND or 3.3v and see if it changes. If ADC value is not changing, then second step is to check if ADC interrupts are called, e.g., by toggling GPIO led in interrupt. If ADC interrupts are not called, you can try to start small: use simplest ADC polling readout function which does not require interrupts and see if it reads ~0 for GND and ~4095 when 3.3v connected to ADC input pin. Idea is to start from simplest working code and move towards more complex design.

@@TechSponsorTV I will try using polling readout to see if it works for multichannel. I have no problem using single adc channel but whenever i switch to multiple channels I have issues. Do u mind if I can mail you?

Good! I usually using CPP program with Sciter.JS GUI to show the data, compile it using CodeBlocks and use custom simple serial port protocol to transfer waveforms to the PC. Thinking about trying Web Serial API. CubeMonitor is fast but a bit clunky, especially on large data. I also surprised that CubeMonitor did not crash even once on my system.

@@TechSponsorTV Yeah I also really like the concept of displaying input data like that graphically. While i'm mostly messing around with STM32 cube IDE atm and STM32 microcontrollers, it appears to be very easy to implement a small HTML server on a raspberry pi pico. It would be interesting to set up measurement data like this which could be accessed anywhere

One approach is to serve data through http server. Data can be fetched using JavaScript from virtual file on IP address. There are some examples which create index.html dynamically and send it through w5500 or esp8266 to web browser. The same way we can make any other data available to the browser and refresh it or fetch using more recent json approaches. It can even work on mobile browser. EDIT: forgot to add that data then can be displayed using HTML5 canvas

I haven't received any emails, please reply here and let me know if you sent anything so I can check - maybe there are some mail delivery problem on my side

@@TechSponsorTV If you still didn't receive my email please let me know

@@TechSponsorTV Just now i have sent an email

Can you please reply to my email

Hey, sorry with the late reply! Unfortunately, I can't help you now. I am swamped with my business and competition. Hope you can find someone on upwork maybe or something like that

for stm32f4 some names are different, here is an example for stm32f4: www.thundertronics.com/STM32CubeIDE/Using_CMSIS_DSP_library.html

@@TechSponsorTV hi sir, i couldnt follow the steps because im using cubemx with the kiel instead of IDE,i hope u can help me

Have you tried this? stm32f4-discovery.net/2014/10/stm32f4-fft-example/ It's pretty old though

@@TechSponsorTV thanks a lot for your help , i have decided to use cube IDE , now im reading and storing sensor data in buffer so,is it the same FFT code u used i should use or there are diffrences?

There are two functions arm_cfft_...() And arm_cplx_mag_...() The first one calculates complex spectrum from complex input data, the second one calculates amplitude (magNITUDE) from complex spectrum. You can use your array as complex input if you interleave it with zeros. arm_cplx_mag_...() function has a design flaw: it loses precision shifting bits left after multiplication or taking a square root.. I don't know where to start, need a better tutorial or one-on-one training. There is also a "real" fft function, but it's not well designed, e.g., it should have same style as complex one, buy instead it's some bogus custom defined function. So, you have some data a[0] a[1] a[2] and so on. Make an array b[k*2]=a[k] b[k*2+1]=0. Feed array b to arm_cplx_fft function, feed output of this function to arm_cplx_mag function... keep array sizes and data types right, visualize each step output to check what's going on. If you need personalized help try reach out to info@thundertronics.com

I think this is the version that comes with STM32 HAL libraries. FatFs version is not the latest too. If there is something interesting about newer CMSIS just let me know, I would be grateful to learn from my viewer!

The simple answer is no. There is some chance of using super-low resolution FMCW (second part of answer) for stationary objects. And for moving objects there is a better option to measure distance: There is a capacitor near FET transistor under metallic shield. If drain-to-source capacitor is removed (or replaced by a smaller value cap), then frequency of this module will respond to supply voltage fast enough to perform FSK modulation. This is the most viable option, because measuring range of ~25 meters will require only 3mhz frequency step, e.g., from F1=24110 MHz to F2=24113 MHz. For modules with quadrature output max. unambiguous distance doubles to ~50 meters z=C/(2*df). C - speed of light, df - FSK modulation frequency step. Possible implementation: some development board with DAC output and ADC input. DAC connected to operational amplifier capable of outputting enough current to CDM324 and constantly outputs meandered signal voltages V1/V2 to achieve switching between two frequencies F1/F2. ADC is connected to CDM324 signal output and synchronized to DAC: samples acquired during DAC low voltage V1 go to array SIGNAL1[k], and during high DAC voltage V2 go to array SIGNAL2[k]. Then, we perform a FFT on both signals and find two complex spectrums SPECTRUM1[k] and SPECTRUM2[k]. We can extract amplitude and phase values for each k-th element of these arrays. E.g., find amplitude spectrum AMP1[k] for SPECTRUM1[k], find amplitude peaks (targets) indexes AMP1[m]. E.g., m=33, 114 and 120 - three targets: slow 33 and two fast 114 and 120. For each target bin "m" we find phases PHASE1 and PHASE2 of SPECTRUM1[ m] and SPECTRUM2[m] using ATAN2 function (google ATAN2). Find difference between phase DPHASE=PHASE1-PHASE2. And convert DPHASE to distance, e.g., -180...0 and 0..180 degrees maps to targets at distances 0..~25 meters moving in different directions (in case of CDM324). For radars with quadrature output unambiguous range 0..~50 meters for a 3MHz FSK modulation bandwidth. Another type of modulation FMCW would be difficult to implement using this module, because it requires much greater tuning range in order of 50MHz or better for usable resolution. It is hard to achieve such wide tuning range, and even if it is possible S21 phase of NE3210S01 and similar FETs used in CDM324 clones may go crazy, and amplitude too. This means that we will have huge variation of amplitude over FMCW sweep which is not good. And frequency may respond too rapid to some tuning voltages - hard to linearize it for FMCW. To sum up, FSK or FMCW can be used after small hardware modification. Unfortunately, FSK can be used only to measure moving targets. There is an option to perform scan of static objects by moving around, so every static object moves relative to the radar, e.g., we can measure distance to roadside trees and other objects if radar moves with a vehicle. But because everything moves, spectrum will have lots of overlapped signals and results may be far from ideal. FMCW: The main problem with FMCW is small achievable modulation bandwidth with CDM324, so range and resolution probably will have ridiculous values for practical use. Btw, older RFBeam radar modules with FSK modulation used exact same approach (and in some cases even same FET transistor), and parallel feedback oscillator is based on very similar microstrip half-wavelength resonator.

@@TechSponsorTV Amazing reply; thanks!!!

@@Jinguapingi Thank you too for letting me know!

Thanks, will do

Thanks for mentioning 3d printed stencils, didn't know this is possible

Thanks for this valuable feedback! I simply liked this music and was kind of selfish making it so loud. I'll try to keep volume lower next time.

You mean impedance mismatch and signal integrity? It is certainly not a problem for chips like STM32. For USB3.0 speeds there will be some "measurable" effect. But it is so negligible. Even at 24GHz if you make 0.5mm microstrip step with line width increased by 0.2mm, you will hardly get any noticeable impedance mismatch or phase shift. In other words, PCB thickness (0.6mm or 1.0mm) will affect performance more than these tiny steps. If you really meant interference. E.g., would this layout radiate more noise. The answer is no, it will not radiate more noise and making some interference. PCB thickness and capacitor placement will have 1000000 times more effect on RF interference. So, you can use it safely in this sense. The main problem with this layout is reliability of PCB assembly. Chip manufacturers have different tolerances for pin sizes shape. If solder paste not touching the pin of certain chip during reflow solder process, it may end up unsoldered. And there is no way to perform visual inspection, because half of solder joints are covered by chip pins. Second problem: such landing pattern is difficult to solder using soldering iron, because each second pad widened area is hidden below the chip. To sum up, I think that 99.999% there will be no any measuramble RF interference increase if this modified landing pattern is used.

It was😨👻

Experimenting with ffmpeg. I've spent few hours to create this monster. A little too bright😂

Thanks! I am trying to find my style. It's not perfect yet, some colors are too bright. Can't wait to make a video on something useful 🤣

@@TechSponsorTV hahahah

What fonts? I may have used Bitstream Vera Sans Mono..

Your music is great! www.youtube.com/@timkuligfreemusic

This video shows how would signal look if we place cheap 24 GHz radar module above the road and drive Tesla CYBERTRUCK under it at a speed of 90 km/h. In other words, it's a simulation of a certain onsite measurement without driving actual car or using a radar. What's most interesting about it is that we can perform simulation using very simple formulas and extract all required data about CYBERTRUCK shape from it's Front and Side photos. How this can be useful? E.g., recognizing vehicles or other objects that are passing near the radar: we can tell if it's a motorcycle, truck or bus. With some more efforts we can estimate vehicle length.

Thansk!

Unfortunately, I haven't saved this sample. You can try to follow this guide: www.thundertronics.com/STM32F103C8T6/Fast_Fourier_Transform.html just go through each step and copy.

I will put more great videos soon!

Links updated, pls check description

@@TechSponsorTV THANKS. HAPPY GETTING RICH!

Most straightforward way is to setup and ADC sampling ~10x frequency of PWM. Something like showed in ruclips.net/video/MBn6LdFENE8/видео.html .If you meant something else let me know and I'll look into it. Maybe a similar but simpler example?

Link in Description updated

Glad you liked it!