EEVblog Electronics Community Forum
Electronics => Projects, Designs, and Technical Stuff => Topic started by: µELKO on November 29, 2024, 07:28:01 am
-
Greetings!
Briefly about me: I am a longtime subscriber to the EEVBlog channel on YouTube, especially in the years of tinkering in my youth Dave's videos have helped me a lot to my current knowledge.
Now I am a development engineer working in the railroad sector (control modules, IO cards).
A recent project with a localization module consisting of an ESP32-WROOM, a UWB module (for indoor navigation) and a ZED-F9R GNSS receiver (RTK) works perfectly as a combined indoor and outdoor localization solution. WLAN was used as the communication interface.
Short information about this module:
- Operating voltage: 12-24 VDC
- 4-layer board (top/bottom signals, layer 2 GND, layer 3 power planes)
- Active GNSS-Antenna (Taoglas AGPSF.36G.07.0100C) stacked on top of the PCB, about 15 mm distance
- No ground plane under the antenna - as it works fine
I then redesigned this module, which resulted in the following desires:
- Communication interface: Ethernet (standard and reliable in the railroad sector)
- Internal power supply: Galvanically isolated
Unfortunately, the GNSS reception on the new board was extremely poor. I found a good comparative indicator, namely the PGA of the receiver (ZED-F9R), to be able to estimate whether the receiver is disturbed even without satellites. Without an antenna, this is 60 dB, i.e. it is fully active. On the old module with WLAN, this is then approx. 42-45 dB with an active GNSS antenna (see picture "Module-WiFi_PGA" -> this spectrogram is reported by the GNSS module itself).
Nevertheless, this image already shows frequency peaks caused by local interference or harmonics. These can either couple in the GHz range of the L1/L2 band or in the intermediate frequency when mixing down - using a radio transmitter, I was able to work out a range of 120-170 MHz in which the ZED-F9R is particularly sensitive. Even sinusoidal frequencies in this range can logically also be seen in the GHz spectrum.
These peaks are even stronger on the new module with Ethernet, which is why the ZED-F9R has to lower the PGA so far in order not to saturate the LNA (see figure “Module-ETH_PGA”). The reason for this is the 50 MHz clock from the ESP32 to the Ethernet PHY, which was routed on the top side of the PCB.
Next, I developed a breakout board of the Ethernet-MAC W5500, which has a local oscillator. This means that no permanent (!) clock signal has to run across the board.
See picture “Module-ETH_W5500-Mod”: Point 2 is the board with the W5500, which was additionally placed under a shielding plate. Unfortunately, this hardly improves the PGA.
All measures to shield anything do not bring any improvement. The current behavior is: When the antenna is screwed onto the soldering studs, the PGA is at its worst. If I set the antenna a little higher, the PGA increases. If the antenna is next to the module, the PGA is good (≥42 dB). And: if the coaxial cable runs over the oscillator of the W5500 MAC, the PGA also drops. So everything is very sensitive.
My guesses so far:
- Issues in the grounding concept. Although there is a solid GND plane (layer 2), this is not designed as a star ground. However, each module (ESP32, GNSS and UWB) is decoupled via ferrites. Sufficient local block capacitors are also available.
- Due to the isolated DC-DC converter, the ground reference floats - but since the GNSS receiver refers to the same, it should not be a problem.
I would appreciate constructive feedback as I would like to understand what is causing this problem. Or starting points as to how I can find out.
Best regards,
Stefan