2
\$\begingroup\$

tl;dr

A tracker device which sends GPRS telemetry has EMI issues. The original design had the antenna connected straight to the "ANT" pin of the GPRS chip. When a "0R" resistor on the feedline path was replaced with a 1pF capacitor, the EMI issues disappeared, and I'm wondering why - is a DC-blocking capacitor actually required?

Background

I've "inherited" a GPS-tracker GPRS-telemetry type of device, which I wrote some extra software for.

The device collects data from an accelerometer, magnetometer, GPS, and records audio. Data is stored on a SD card. Also, twice per minute, an aggregated packet of telemetry is sent over the GPRS link to a server in a fire-and-forget style (UDP).

The hardware design of the device is not stellar, but looks good enough. The GPRS chip (ublox Sara G350) is mounted on a separate 2-layer PCB with a lot of ground plane on both sides, heavily stitched. The GPRS antenna feedline is short and is a coplanar waveguide with ground, correctly dimensioned for 50Ω impedance. There's more than enough bypass capacitance on the GPRS chip VCC, my only gripe is that the small nF/pF caps could have been closer to the VCC pins.

"hiccups"

During testing it was discovered that the device has a latent EMI problem, which was causing instability and corrupted data: when operated in an area with poor network coverage, on each 30-second interval (i.e. coinciding with the GPRS transmit), one can see the following effects:

  • battery voltage dips slightly (which is expected and not EMI-related - just a result of the GPRS chip pulling a lot of power);
  • input power voltage measurement sees phantom voltage spikes of a few millivolts;
  • accelerometer samples contain garbage for a few 10s of ms;
  • the audio chip locks up and requires a reinit (rarely);
  • the main MCU resets without a reason (very rare)

I'm calling these tell-tale signs "hiccups", because they occur spuriously, and everything is otherwise fine in between them.

The core EMI issue might be something related to the I²C bus, as sensors not on the I²C like the GPS seem unaffected. But I'll ask another question for it after I understand it more, as it would be too broad right now.

The devices do not hiccup in my office, but if I take the device and drive it on a mountain road, where the network reception is known to be bad, you can later review the stored telemetry in the SD card and see how the hiccups appear and increase in intensity as the signal gets worse and worse. The GPRS chip also reports RSSI. The hiccups and RSSI are well correlated.

Matching circuit

The device can use either an internal GPRS antenna (PCB-type), or an external antenna on a SMA connector. There are footprints for a π filter between the GPRS chip and the antenna. After the filter, there's also a footprint for a U.FL connector, used for the external antenna configuration:

  • internal config: matching circuit passives are populated, U.FL connector is not populated, the antenna (Antenova A10340H) is soldered as per datasheet requirements;
  • external config: matching circuit is just a 0R; feedline is cut from the U.FL onwards; the U.FL connector is populated, a short coaxial cable leads to the SMA connector outside the enclosure. The Molex 2144290001 antenna is used.

I initially looked into the "internal antenna" devices, and noticed that the matching circuit was different from what Antenova recommended. We changed it to match their reference circuit, and the hiccups practically disappeared.

The "external antenna" devices were less hiccupy from the start, but still had issues, and the following sentence from the Antenova datasheet caught my eye (emphasis mine):

dc blocking cap requirement

I pored through the GPRS chip datasheet to see if they too recommend a DC blocking cap, but found nothing. Nevertheless, I decided to test and replaced the 0R series 0603 resistor of the π circuit to a 1pF 0603 C0G capacitor. What I saw was:

  • the hiccups virtually disappeared;
  • the reported GPRS RSSI was lower by some 5-10%;
  • packet loss between the device and the server was comparable (since it is UDP, if there's no network coverage, the data is lost).

The comparison between the two configurations (0R vs 1pF) uses the same device, mounted on the same car, the same antenna orientation, diven along the same route through a mountain road with known bad reception in certain areas. The only difference should be the series element of the π filter. I've also replicated these results using another device with external antenna, which again went through the 0R→1pF replacement.

Questions

  • is a DC-blocking capacitor really required, as the documentation (Molex antenna datasheet and the u-blox Sara G3xx system integrator's manual) mentions nothing in that regard?
  • what could be the reason that adding the capacitor reduced the EMI issue?
\$\endgroup\$
6
  • \$\begingroup\$ it makes sense that it would generally require a blocking capacitor, but the symptoms seem unusual - is it possible that your supply is dipping more than you think? What is your sampling rate for the "slight dip" that you measure? What type of antenna do you have? If it connects to ground, the chip may draw too much current because of DC stuff \$\endgroup\$
    – BeB00
    Commented Feb 22, 2023 at 8:07
  • \$\begingroup\$ The GPRS PCB includes 6×330µF capacitors in parallel to address the supply issue. I've also tried including more, parallel to the battery, and that did not improve anything. The sampling rate for the slight dip is 5 Hz. The antenna models are listed in the question. I do not know what their construction is, though. \$\endgroup\$
    – anrieff
    Commented Feb 22, 2023 at 9:25
  • \$\begingroup\$ Mountains are preferred transmitting sites for broadcast, utility, civil communication. Your receiver may have difficulty discerning weak signals against this huge background - an antenna matching network might help attenuate out-of-band signals, as does your 1pF "network". However, this doesn't explain lockups, accel-garbage, etc. A bad antenna match can result in large transmitting RF currents back inside your electronics - creating digital & analog havoc. Changing (improving) RF ground may make a difference. \$\endgroup\$
    – glen_geek
    Commented Feb 22, 2023 at 13:49
  • \$\begingroup\$ I would suggest monitoring the power rails with an oscillocope, right next to the device \$\endgroup\$
    – BeB00
    Commented Feb 22, 2023 at 21:44
  • \$\begingroup\$ "...and noticed that the matching circuit was different from what Antenova recommended. We changed it to match their reference circuit, and the hiccups practically disappeared." Well this was obviously the problem then, more so than the DC blocking cap. All chip antennas require matching. Unless you are using an antenna that the module manufacturer explicitly recommend to use "drop-in", expect that you will need matching. \$\endgroup\$
    – Lundin
    Commented Feb 23, 2023 at 9:50

2 Answers 2

1
\$\begingroup\$

It would be worth asking Antenova why they require a DC blocking capacitor; I'd suspect it is just a precautionary measure in case the GSM modem doesn't include DC blocking.

Maybe the 1pf capacitor is acting as a high-pass filter, attenuating the 900 MHz signal sufficiently that the unit only connects at 1800 MHz, and the maximum permitted transmit power at that frequency is less, which results in lower PSU current, and hence fewer dropouts on the supply line.

Regardless of the reason, your system does have a problem handling high GSM transmit powers, maybe because the PSU drops out, or the circuitry is picking up the RF signal, and this does need to be addressed.

One way to quantify the problem is to get hold of a desktop RF screen box (that is generally available second-hand for a reasonable price), and equip it with internal & external GSM antennas, linked via a step RF attenuator. You can then check the effect of a reduced RF signal, and any circuit improvements, in a more controlled environment.

\$\endgroup\$
2
  • \$\begingroup\$ Thank you for the RF screen box / RF attenuator idea. It would be indeed useful to help debug the root issue. I'm inclined to think it is real a EMI issue, not a power supply one. The reason is that the GPRS chip has well-described power transients spec, requirements on power supply ESR, etc., and these seem to be all met. In fact, the GPRS chip is powered directly from the Li-Ion battery, while all other sensors and chips are from a 3.3V rail. Even with 2A bursts, the Vbatt rail would dip by some 200-300mV (down from 4.2V), so the 3.3V rail should be steady, as the LDO has good PSRR [...] \$\endgroup\$
    – anrieff
    Commented Feb 24, 2023 at 22:23
  • \$\begingroup\$ [...] I have indeed tested how the device would perform at low battery, less than 3.3V, where the LDO does not regulate anything and any GPRS-induced dips are also present in the 3.3V main rail. Tests however show that, if anything, the device performs better in this case, I assume because of the reduced Tx power. \$\endgroup\$
    – anrieff
    Commented Feb 24, 2023 at 22:25
0
\$\begingroup\$

“is a DC-blocking capacitor actually required?”

If there are sources of DCDC converter noise or high E fields coupled onto the ground or antenna cable, this will affect performance for low S/N ratio signals.

If a 1pF cap at -j100 ohm at 1.5GHz to antenna appears to help, then it appears that blocking lower RF signals explains the result, rather calling it DC blocking.

\$\endgroup\$
2
  • \$\begingroup\$ Thank you for the pointer, yet I don't think DC-DC converter noise is the culprit. The device indeed has a DC-DC for battery charging, but it's optional. If there's no input power, the DC-DC is off. The hiccups still occur in that scenario... \$\endgroup\$
    – anrieff
    Commented Feb 22, 2023 at 14:41
  • \$\begingroup\$ do a test in office with something to attenuate the signal . Look for RSSI signal to show AM EMI or put spectrum analyzer with loop wire to sniff EMI spectrum inside Rx to spot potential EMI , same ac coupled DC. .investigate all sources of EMI at low RF \$\endgroup\$
    – Hoagie
    Commented Feb 22, 2023 at 16:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.