Hi Team,
My name is Abbas Ali. I am a PhD researcher at the University of Michigan-Dearborn, working in the Information Systems, Security, and Forensics (ISSF) Laboratory. Our lab focuses on hardware fingerprinting and security of embedded systems and automotive ECUs.
We are seriously considering purchasing the RFSoC 4x2 for our research and I had a few technical questions before we move forward.
Our primary use case is acquiring analog signals directly from automotive ECUs — specifically CAN_H and CAN_L signals. CAN_H typically sits between 2.5V and 3.5V, and CAN_L between 1.5V and 2.5V. We also work with 5V GPIO signals from microcontroller boards like Arduino. After reviewing the XCZU48DR datasheet, I noticed the ADC_VIN absolute maximum rating for ZU4xDR devices is 2.1V, which means our CAN bus signals would exceed this limit.
So my questions are:
Is there any onboard signal conditioning or input protection on the RFSoC 4x2 that would allow us to safely feed CAN bus or 5V signals into the ADC inputs, or are the SMA inputs directly connected to the ADC pins with no protection?
If signal conditioning is required, do you offer any recommended or off-the-shelf conditioning circuit that works well with the RFSoC 4x2, or would we need to design our own?
In parallel with data acquisition, our longer term plan is to deploy machine learning models directly on the FPGA — specifically transformers and reinforcement learning algorithms for real-time device identification. We confirmed that the RFSoC 4x2 supports DPU B4096 with 2 cores via DPU-PYNQ. Is this configuration stable and well tested for inference workloads, or are there known limitations we should be aware of?
We are genuinely excited about this platform and are planning to purchase some boards for the lab. Any guidance you can offer would be really helpful before we finalize the decision.
Thank you for your time.
Best regards,
Abbas Ali
Ph.D. Student, Electrical and Computer Engineering
University of Michigan-Dearborn
Hi @mrabbas, welcome to the PYNQ community.
Exceeding the the RFSoC’s ADCs maximum input voltage will likely damage the device and is not recommended. Additionally, while the RF-ADCs have a max voltage you shouldn’t be near that level most of the time. There is a normal voltage range you should actually adhere to instead. You’ll find more info in the device datasheet and PG269 data converter user guide.
As you suggested, you will need to condition the signal before connecting it to the RF-ADCs. Unfortunately I don’t know of any COTS part that will do this off-hand, but you may find some level translators online with the correct spec for what you need.
Hi @joshgoldsmith ,
Thanks for your response. I will look into the signal conditioning solution, and will check the datasheet for the normal voltage range.
But in parallel with data acquisition, our longer term plan is to deploy machine learning models directly on the FPGA, specifically transformers and reinforcement learning algorithms for real-time device identification. We confirmed that the RFSoC 4x2 supports DPU B4096 with 2 cores via DPU-PYNQ. Is this configuration stable and well tested for inference workloads, or are there known limitations we should be aware of?
Thanks.
Regards,
Abbas Ali
Hi,
One additional important question from our side:
Would it be possible for you to share an example or reference of a square-wave capture using the RFSoC 4x2 ADCs? We are particularly interested in seeing the expected waveform quality from the platform before moving forward with the purchase.
For our research, we are not only interested in the ideal square wave itself, but more importantly the rising/falling edge transitions, ringing, ripple, overshoot, and settling behavior after the transitions. These transient characteristics are very important for our hardware fingerprinting research.
So we would really appreciate if you could share:
-
An example square-wave capture from RFSoC 4x2
-
Any screenshots from Oscilloscope or other acquisition tools
-
Any documentation, website, demo, or reference resource that shows real captured waveforms from the RFSoC 4x2 platform
Seeing the actual expected waveform behavior would help us better evaluate whether the platform is suitable for our signal acquisition requirements before purchasing multiple boards for the lab.
Thank you again for the guidance and support.