because it came up recently as a support question and we’re working on getting the rest of the batch of LS1028A modules shipped, I successfully tried to reproduce reports of WiFi causing display glitches on the MNT Reform eDP display and made some videos that show what’s going on.
Also, we have a revised eDP adapter for LS1028A that flips the pinout of the FPC connector, so we can use a shorter, contacts-on-same-side flat cable that doesn’t require origami. This will be included in any LS1028A shipments going forward and we will send it free of charge to anyone who already has their module and has display glitches even without WiFi. To claim this, just write to firstname.lastname@example.org with your order number and reference this post.
My findings after a few hours of interference experiments today:
- It depends on WiFi frequency. For me it doesn’t happen with 5GHz, only with 2.4GHz bands.
- It depends on antenna position. If the antenna is close to the part of the display cable that connects to the display adapter (nonsleeved segment), it is especially pronounced.
- Unfortunately with LS1028A and the first-gen acrylic antenna holder, the antenna is in the worst possible place.
Some quick workarounds until we have put together a different antenna/acrylic bundle:
- Use 5GHz WiFi if possible.
- Find another temporary location for the antenna–for example, tape it to the inside of the acrylic bottom plate further away from the eDP cable.
- Use a WiFi antenna with a longer cable or use an extension cable to feed it to the other side of the case.
- Use a WiFi USB stick.
- Possibly tape the unsleeved part of the display cable (with the little seperate wires) with a layer of metal tape and another layer of isolating plastic tape. This is a bit messy though.
Here are 2 videos showing the effect (with two different antennas):
In my testing I get occasional interference with WiFi and none when WiFi is off (wired Ethernet only). I suspect the WiFi is doing something special from time to time that causes this interference. I do not know if I use 2GHz or 5GHz, I have a network with same SSID and password in both bands.
EDIT: I removed the wifi module and the display looks fine (no glitching). To me, this is pretty concrete proof that the wifi card was the culprit. I should have checked that right before posting instead of right after. I’m just going to get an external wifi adapter and stick with the rest as-is I think.
I think I have the same-sided cable mentioned above (did not require origami to install) and I still see visual display glitches.
I have the wifi module installed but taped the antenna acrylic next to the trackball holder, so is decently far away from the cable.
I don’t really have a good way to measure the signals on the display cable at home, and I’m not sure what the visual glitching is correlated with. This happens on first boot with a fresh image, and while a wifi card is installed, I haven’t told it to do anything or connect, so not sure the wifi is active.
One hypothesis is I bent the cable too much while installing. The exit from the SoM connector is pretty tight, and possible it’s bent a little too much there. Also might have bent the cable in other places.
[EDITED to remove hypothesis about running without batteries]
I think my biggest takeaway is that I wish I had some device that could see/measure the display signals at the display adapter, since that would help me understand what’s going on better, I think.
An update: As we’re shipping some more backlogged LS1028A devices I’ve looked into this some more. First, we replaced the Molex antenna with one that has 20cm cables and used a new acrylic (from the RCM4 project) that mounts over the SD card slot, so the antenna is at the other end of the device. But I still got interference into eDP on 2.4GHz WiFi (AFAIK not on 5GHz, so one workaround is to disable 2.4GHz).
My working theory is that the pinout of the eDP FPC connector on the LS module combined with a simple flat cable is bad for signal integrity, as there is no ground plane and no managed impedance between the N/P lines of each lane:
As you can see, N/P pairs have a GND pin in the middle, and TX0_DN is closer to TX1_DP than to TX0_DP. It should be the other way around.
So I tried to design a custom flex PCB (cable) that fixes these things:
- Diff pairs are routed together with the right impedance
- Diff pairs, not their lines, are separated by GND
- There’s a GND plane on the bottom layer
- EMI shielding foil will cover the cable and will be connected to GND lines
I’ve ordered 20x at JLC and hope that they will arrive quickly after CNY and will report back with testing results.
We will continue shipping LS1028A Reforms as-is to not let everyone wait more, and if the flex cable works out in tests, mail these to anyone who needs for free.
I wonder if the length of the cable is part of the problem.
1/2 wavelength at 2.4ghz is between 6 and 6.25cm depending on what channel. Close to that and you might have enough voltage at the ends to interfere?
How would 2.4GHz be disabled on a dual band card?
It seems that even on 5GHz network the card would regularly tune to 2.4GHz, at least that would explain the observed behavior with the card enabled.