I recently designed a PCB that contained a buck regulator (LM2576 5V). You can see the PCB layout in this question. After converting the input 12VDC to 5V, two linear regulators then bring that down to 3.3V (for a micro) and 4.1V (for a SIM800C GSM modem).
When the board is connected to other boards in the system (which contains a speaker), a loud GSM buzz is heard when the modem transmits. Bringing a mobile phone close to the speaker board also caused the buzz but it was much much quieter, which implied to me that the noise was being conducted through the cables, rather than being radiated.
In order to understand the noise, I did a simple conducted emissions test (using a spectrum analyzer and a Tekbox LISN). I captured the conducted emissions from 150kHz to 30Mhz (RBW 9kHz, step 5.4kHz dwell time 50ms) in a variety of different firmware configurations to try and understand where the problem originates from. Here are the scans:
Scan 1 with GSM modem transmitting as frequently as possible (roughly once every 750ms) The large spikes spread throughout the spectrum are usually ~300kHz apart (closer peaks are about 90kHz apart - not sure how to interpret that. Could be multiples of the 52kHz switching frequency.)
However, I then decided to solder a 4.7R power resistor to the output of the 4.1V regulator to simulate the GSM modem load continuously.
The overall noise jumps back up again to similar levels to the spikes that were present in the first trace with the GSM modem on. If I were to join the peaks of the spikes on Scan 1, I'd end up with a trace similar to Scan 3. This made me think that the problem may actually be the SMPS generating more noise at higher loads rather than it be noise caused by the GSM signal itself, and that the GSM noise was actually due to the modem demanding sudden bursts of power from the SMPS which was then working harder to provide that and generating lots of switching noise.
If this is true, it surprises me for the following reasons:
- I tried to follow the recommended layout for the LM2576 as specified in the datasheet to minimize switching noise.
- The 4.1V regulator has both large and small value capacitors on the input and output, which I would have thought should be sufficient to smooth out the current requirements of the GSM modem.
- The GSM modem has 100uF tantalum, 1uF, 33pF and 10pF caps positioned close to the power supply pins, as recommended in the datasheet, which should handle the high current requirements of the modem.
- Although there are layout issues with the board (especially underneath the micro in the center), the power supply traces seem OK in terms of current loops (see the layout here).
The LM2576 datasheet makes no mention of any sort of input filter before the SMPS (only output filters).
So the questions I'm asking are the following:
- Is my analysis of the source of this EMC problem correct? (SMPS generating input noise due to high load).
- Are input filters always necessary in order to reduce SMPS switching noise, or can this problem be solved just by improving the PCB layout? If so, where are the layout problems that are causing this issue?