Reducing power supply ripple for GSM modem

I'm powering a SM5100B GSM modem from a 5A bench supply (peak current is 2A). The required supply range is Vmin=3.3V, Vnom=3.6V, Vmax=4.2V, so I'm giving it 4V. Every so often, the module will reboot (a problem I've had with all three GSM modules I've tried: this, SIM908 and ADH8066). I presume that it's a power issue. My power is not dropping below 3.3V, but I saw in the SM5100B data sheet that the maximum ripple is 50mV$_{PP}$ for freq <200kHz and 2mV$_{PP}$ for freq >200kHz. I have a ripple that's generally around 40mV, but it does spike upwards sometimes. I put it on my DSO in pass/fail mode with a 50mV$_{PP}$ pass rule:

I get a chirp every two seconds or so, and they look to spike up to about 100mV or so. I thought I might be able to fix the problem with a couple of capacitors, so I added a couple of 16V 1000$\mu$F electrolytics across the power supply rails on the breadboard, close to the GSM module. Unfortunately, it doesn't seem to have made the slightest bit of difference. What else can I do? Is the ripple really likely to be the problem, or am I chasing a red herring?

• Try smaller capacitors: small ceramics of 100 nF or maybe smaller. Low inductance and ESR is the key. Also remember that you don't really have an RC filter without sufficient R. When capacitors are used to filter ripple, it's with the understanding that wires and traces provide the L and R. But if that is too small compared to the C, then it doesn't drop much voltage. – Kaz Jan 6 '14 at 21:35
• How are you probing...are you using a very short ground lead? – dext0rb Jan 6 '14 at 22:48
• I'm now using the jack on the eval board, which goes through the on-board LDO regulator, but that hasn't helped. It will actually run and allow me to send messages over GPRS continuously for a couple of minutes before it resets itself, so whatever problem there is seems to be very close to tolerance. – Isvara Jan 12 '14 at 4:10
• 2 sec peaks is standard GSM tower communication. There the module will draw a bit more than 2amps for a very short period of time. Instead of a PSU, use a low internal resistance LiPo battery to power the module directly. (If you still need to run it from wall power, I'd suggest a dedicated 3.3V step down regulator with low esr capacitors as others have suggested. I have had endless issues with bench PSUs and thin bench cables.) – Gee Bee Mar 17 '16 at 0:45

It is more than likely that what you see on your scope, is not the problem you are chasing.

From the description, my first suspect would be the breadboard itself. Breadboards are notorious for having extremely bad contact resistances (like, 1 ohm, or similar), which may be of good service to provide power separation BUT only in case you have proper bypass capacitors on the load side.

From the documentation of the GSM module, it looks like it has a small, fine-pitched SMD PCB connector. So, there must be some kind of interconnecting board this module is plugged into, and that board should host your low-ESR 470-2200uF electrolytic capacitor (as close to the connector as possible), not the breadboard.

• Using the jack on the dev board now -- see update. – Isvara Jan 12 '14 at 4:11
• I'm now powering the modem directly from the output of the DC-DC converter instead of from the breadboard's power rails, and I've gone about ten minutes so far without a reset. If it survives soak testing, I'll accept this as the answer. – Isvara Jan 15 '14 at 6:15

I would guess that to be a red herring. You're looking at a signal on the order of 100 MHz, with very low amplitude. I'd suspect that's actually some local radio station you're seeing on your scope, and not real noise. You can help confirm this by putting both ends of the scope probe on the common node of your PCB. If the noise is still showing up on the screen, it's not real.

I'd suggest altering your input voltage. If Vnom is 3.6, give it 3.6 and see if the problem goes away.

• Okay, you're right, the noise still shows up when I do that, so it looks like ripple is not the issue. Also, see my update. – Isvara Jan 12 '14 at 4:10
1. Are you sure this is PSU ripple? As Stephen said, the frequency is around 100 MHz, a bit high even for a switching PSU. One way to test that is to increase/decrease the load. Another way is to feed the modems from batteries (zero ripple); two lead-acid cells should do the trick given your current requirement.
2. Even if it is, you need some impedance "before" the capacitor to complete the filter - may I suggest an RF choke?