Blind spot monitoring unit teardown!

[kgavionics]

Hello guys!
I just wanted to share my teardown of a defective Blind spot monitoring unit! This unit was installed on Mazda CX-9 2011.
I want to know more information about how the system works and any useful information about it!
Thanks in advance

[MisterHeadache]

The third pic looks like a second board.  What's on its other side?  The side you show sure looks like phase array radar transmit and receive antennas.  The three round cross shaped soldered objects are MMIC amplifiers.  There - I have exhausted my knowledge of what I see here - others who deal with this stuff every day surely can add more detail.

[kgavionics]

I attached the bottom picture of the antenna!

[Georgy.Moshkin]

It has three receiving antennas (with three MMICs or microwave FET amplifiers). Gate pin of each amplifier is biased using thin feeding line with two radial stubs and a resistor. Antennas RX1/RX2 are for azimuth (horizontal) angle sensing when TX-A is transmitting . Antenna RX3 is most likely only for long range distance sensing when TX-B is transmitting. Theoretically, RX3 can be used together with RX1 or RX2 to measure elevation (vertical) angle. Probably not doing any elevation angle measurements here. Transmitting antenna TX-B is narrow-beam high-gain for long-range distance measurement mode (when RX3 is receiving). In this mode, it work as simple fsk/fmcw radar with single TX / single RX  antenna. Transmitting antenna TX-A is wide-beam low-gain for short-range distance AND angle measurements (when RX1 and RX2 are receiving). Angle measurement is usually done using phase monopulse method (unambiguous angle measurement range depends on antenna center spacing d1/d2) In some older designs, the same MMIC/FET is used as active device for oscillator (under shield on the back side). In my design I also used FET as resistive mixer, but it is kind if rare. More interesting stuff may be under the shield on the back of the antenna.

[Georgy.Moshkin]

DSP/control board contains bunch of operational amplifiers. Depending on MCU/ADC performance and modulation type, there are few options how it can be made. Microwave module has inputs and outputs: power, vco control, some other control inputs. Outputs: signals from each antenna's receiving mixer. Those outputs are fed to operational amplifiers, and then amplified and filtered (low pass+high pass) signal goes to multi-channel ADC. MCU prepares digitized data from each channel to go through FFT algorithm. E.g., if it's FSK radar, MCU can control VCO using GPIO or DAC. This process is synchronized with ADC reception, so for TWO fsk frequencies we can extract TWO signals from ONE antenna. Using signal from single antenna is already enough to measure distance (e.g. digitized 2048 samples, made two 1024 arrays A1 A2, performed FFT1 FFT2, found peak value at element N, calculated phase difference between FFT1 and FFT2 at element N. Used formula to convert N to Hz (Doppler) and then to speed (km/h). Used another formula to convert phase difference FFT1[N]-FFT2[N] to distance). Perform the same for second antenna (FFT3 and FFT4). It will provide the same speed and distance, and also target angle can be extracted using FFT phase of adjacent antennas FFT[1]-FFT[3] or FFT[2]-FFT[4]. E.g., on STM32H7 I have 8 channel FFT extraction of all peaks and angles for my "Radar for the blind and the deaf project", with sampling time offset compensation, sliding window and other for best performance. On STM32F1, it is possible to process radar signals like on your photo, but ADC may be a limitation on F1, but can be solved using analog sample-and-hold circuit (opamp + spdt + 2 X capacitors + 2 X opamps). This circuit will extract two copies of 1024 data points for FSK f1 and f2 carrier frequencies analog way. It doubles number of required ADC channels but lowers sampling rate frequency down to few kHz instead of MHz. And the trick is to use CMSIS FFT followed by fast optimized ATAN2 phase difference extraction. I use ATAN2 based on ATAN lookup table. Many radars use FMCW, and it is pretty simple too. Need to distinguish between "slow" and "fast" FMCW. Radar on the photo is 24GHz (or near) band. Such radars use "slow" FMCW, and some use MFSK (FSK superimposed on FMCW ramp). Newer 77GHz  radars usually use "fast" FMCW. But there are more mofulations and algorithms, e.g. "MUSIC" and others. Just Google for those. So where to start? Find out power pins, control pins and mixer output pins, and you can start experimenting by observing signals on oscilloscope or microcontroller, and try different modulation, FFT, etc. If DSP board is not severely damaged, you can reuse it's opamp part, but it will require good soldering skills. I'll be checking this thread! Thanks for the photos, I collect them.

Edit: probably most important thing that can help is to experimenting with recorded signals.  E.g., you can record two channels from two antennas using stereo microphone input, and then try different algorithms using Matlab, Python, custom cpp wav reader with fft, peak finding, extract target speed and angular position.

[kgavionics]

Thank you, Gregory, for this awesome explanation !

[Georgy.Moshkin]

Blind spot monitoring unit on your photos may have similar microwave circuit to my 2017's design, which is a 10GHz radar that can measure speed, angle and distance of moving objects. I used cheap FR4 substrate, but microwave part worked surprisingly well despite dielectric constant variations. I think it's because aperture coupled patch antennas are wideband. In this design, I tried to use oscillator idea from patents US6064276 / US5262783. Project was abandoned, but I learned a lot from it. Here are some comments on my photos:
1. Oscillator (in my case parallel feedback oscillator) signal amplified and divided to two paths: TX (goes to TX antenna) and RX ("LO" signal source for receiving mixers)
2. TX signal is radiated through TX antenna array
3. Signal from each RX antenna is amplified using "RX AMP" and mixed with TX("LO") signal using it's own mixer (resistive fet mixer in my case)
4. Frequency is modulated using coupled stub (gate pin of FET is biased using DAC signal from STM32F4)
On the other side there is DSP/control board:
1. Signals from antenna mixers (RX1, RX2) are going to preliminary amplification 2-channel OPAMP. This OPAMP is needed to provide enough current for charging 4 "sample and hold" capacitors.
2. Amplified signal RX1 goes to SPDT1. This switch is synchronized with STM32's DAC and GPIO output (output compare,etc.).
3. RX1 voltage level during FSK modulation is "saved" using two capacitors on SPDT1 output and fed to next OPAMP. This is where single analog signal RX1 is  separated to two analog signals RX1 "A" and RX1 "B", as if we had two radars with frequencies Fa and Fb transmitting simultaneously
4. The same operation of signal amplification/separation is performed on RX2 channel (capacitors 3,4 on SPDT2 output)
5. Amplified signals are fed to I2S stereo ADCs (I've used PCM1208 audio ADCs). That gives 4 channels in total

EDIT: In other words, all those blind spot monitor radars are using the same principle:
They always have have one or more transmitting (TX) antennas. Multiple TX antennas used for different purposes: beam steering, switching between different radiation coverages.
They always have one or more receiving antennas (RX). In most cases, multiple RX antennas are used for phase monopulse angle measurement.  The rest of it is mathematics: we know what was radiated, what is being received, and what data can be extracted from it. Accompanying analog and digital circuits are used to control TX oscillator and to perform downconverted RF signals conditioning and processing.

 BTW, if you search on GitHub, you can find communication protocol descriptions of some collision avoidance radars, and I think it's possible to buy used unit somewhere to do experiments. I can't find a link now, but I am pretty sure there was several CAN drivers for such radars, where you can set different control registers and read distance and speed measurements. I talked once to a seller of such modules, and he had no clue if they were working or not. Less than $50 for a dusty 70GHz unit with some barcodes on it with no information online. So I didn't bought it.

Check this link, it demonstrates well how communication with such devices may ook like:
https://github.com/VT-ASIM-LAB/continental_radar_driver

[MarkS]

RF engineering is witchcraft, I swear!