[Help] Optimising decoupling via transient load sim

[BadeBhaiya]

Hello all.

tl;dr need guidance, best way to optimise decoupling for GSM module. I have recorded the actual transient load profile of the module and am using it in simulation. Help me make the simulation as close to reality as possible

Long story

I am updating an existing board that my company uses. It contains a GSM module and a few other 2.4GHz radios. We had been facing issues because the previous designer omitted on-board decoupling and relied only on a 20F supercap in parallel with the 3.6V battery (a Li-SOCL₂ cell). The supercap in question was from a dubious supplier, and its ESR greatly increases when exposed to cold (-5C max) but it never recovers when the weather warms up again. Because of this, the device fails to transmit since it fails to deliver the 2.5A peak the GSM module needs, without dropping more than 0.2V on the supply rail.

I have used an NRF-PPK2 load profiler device to capture the transient load profile of the GSM board, and made an ltspice circuit with the recorded load and approximate trace parasitic elements in an attempt to get the perfect amount of decoupling required. If possible, I would also like to current limit the battery to hopefully decrease the high peak current and extend it life. Best leave this for later though.

I have a rough idea that electrolytic caps are out of the question, because of their temperature sensitivity so I will be using tantalum or X5R/X7R. Aluminium polymer are on the table, but they can get expensive. I can even add a step up switcher to maintain a steady voltage to the GSM module.

Is there anything that I can add to the simulation model to make it more realistic? Or is it as simple as adding more caps until it starts working?

[mtwieg]

Sounds like simply replacing the supercap would be the simplest fix, probably by a wide margin. Any other type of capacitor is going to require far more board space and likely be much more expensive.

What's the maximum duration and duty cycle of the load transients? How much peak current can the battery supply? This will determine whether capacitance or ESR is the limiting factor for you, and thus what capacitor type is most suitable (probably EDLC).

If possible, I would also like to current limit the battery to hopefully decrease the high peak current and extend it life. Best leave this for later though.

This will require more decoupling capacitance, of course.

Is there anything that I can add to the simulation model to make it more realistic? Or is it as simple as adding more caps until it starts working?

If anything, this model is overcomplicated. Nanohenries and picofarads aren't going to be relevant for load transients like this. Of course they are relevant for high frequency decoupling, but that will not be taken care of by your bulk decoupling caps. IMO its best to ignore such small caps entirely for this simulation.

[moffy]

Plastic film are relatively stable and might be added to the mix of other caps, polypropylene have nice HF characteristics.

[BadeBhaiya]

Sounds like simply replacing the supercap would be the simplest fix, probably by a wide margin. Any other type of capacitor is going to require far more board space and likely be much more expensive.

What's the maximum duration and duty cycle of the load transients? How much peak current can the battery supply? This will determine whether capacitance or ESR is the limiting factor for you, and thus what capacitor type is most suitable (probably EDLC).

I tend to agree with you, the issue however is that the supercaps are EDLC type, which are quite sensitive to temperature changes (they work fine above 0C but the ESR increases exponentially below that).

The transient current waveform is the I1 in the simulation which is feeding data from a ~70MB PWL file. The average current is ~50mA, but pulsed current can touch 600mA and above. Nothing too crazy though, compared to what the GSM module's datasheet claims (2.5A peaks). Its neigh impossible to put enough decoupling to deal with a failing supercap or a high internal resistance cell alone.



Because of this, and board space being a consideration, I am leaning towards having a switching regulator on board to actively maintain the voltage to the GSM module. Its the only way I can think of, which will work out with board space, and without putting a truck full of tantalum caps.
Testing with an off the shelf evaluation board for a switcher (MAX18000) has given me good results so I'm optimistic. I'll use something similar I guess

If anything, this model is overcomplicated. Nanohenries and picofarads aren't going to be relevant for load transients like this. Of course they are relevant for high frequency decoupling, but that will not be taken care of by your bulk decoupling caps. IMO its best to ignore such small caps entirely for this simulation.

Noted, makes sense. Its just a random model that I thought up 

Plastic film are relatively stable and might be added to the mix of other caps, polypropylene have nice HF characteristics.

I'm not too familiar with plastic film types, but aren't they for high voltage applications? Like in mains stuff where a few nanofarads worth of capacitance is needed at 600V or whatever?

[AnalogTodd]

Sounds like simply replacing the supercap would be the simplest fix, probably by a wide margin. Any other type of capacitor is going to require far more board space and likely be much more expensive.

What's the maximum duration and duty cycle of the load transients? How much peak current can the battery supply? This will determine whether capacitance or ESR is the limiting factor for you, and thus what capacitor type is most suitable (probably EDLC).

I tend to agree with you, the issue however is that the supercaps are EDLC type, which are quite sensitive to temperature changes (they work fine above 0C but the ESR increases exponentially below that).

The transient current waveform is the I1 in the simulation which is feeding data from a ~70MB PWL file. The average current is ~50mA, but pulsed current can touch 600mA and above. Nothing too crazy though, compared to what the GSM module's datasheet claims (2.5A peaks). Its neigh impossible to put enough decoupling to deal with a failing supercap or a high internal resistance cell alone.

(Attachment Link)

Because of this, and board space being a consideration, I am leaning towards having a switching regulator on board to actively maintain the voltage to the GSM module. Its the only way I can think of, which will work out with board space, and without putting a truck full of tantalum caps.
Testing with an off the shelf evaluation board for a switcher (MAX18000) has given me good results so I'm optimistic. I'll use something similar I guess

If anything, this model is overcomplicated. Nanohenries and picofarads aren't going to be relevant for load transients like this. Of course they are relevant for high frequency decoupling, but that will not be taken care of by your bulk decoupling caps. IMO its best to ignore such small caps entirely for this simulation.

Noted, makes sense. Its just a random model that I thought up 

Plastic film are relatively stable and might be added to the mix of other caps, polypropylene have nice HF characteristics.

I'm not too familiar with plastic film types, but aren't they for high voltage applications? Like in mains stuff where a few nanofarads worth of capacitance is needed at 600V or whatever?

You're looking at some fairly short duration pulses in there that require significant amounts of current. What needs to be thought about is the impedance a capacitor presents at the speeds you are looking at. If your load current pulse took 100msec to rise and fall, the supercap or bulk capacitors you have in there would be fine. It goes back to I = C * dV /dt, where the I required in that case is fairly low to keep dV/dt low. With faster pulses, you now see higher dV/dt as the I gets a lot higher and this effect is multiplied with the ESR of the capacitors. Think about it--your trace parasitics are all in the pF range, they can't hold enough charge to do anything. Your bulk capacitors have 1ohm of ESR, so if each one is delivering 250mA pulses, then they will drop 250mV each just across the ESR.

You want capacitance that has low enough ESR that you don't get these transient voltages for fast load currents and large enough that they store enough charge to hold up the output during the load current transient. I'd be willing to bet that the evaluation board you got has some reasonable MLCC capacitors on it--low ESR and overall impedance out to high frequency. You want a mix of capacitors on your board--small ones with low ESR to deal with high frequency transients (usually MLCC) and some larger size with higher ESR to deal with holding up from a bulk (DC) perspective.

[mtwieg]

I tend to agree with you, the issue however is that the supercaps are EDLC type, which are quite sensitive to temperature changes (they work fine above 0C but the ESR increases exponentially below that).

Some EDLCs/supercaps tolerate freezing temperatures much better than others (depends what kind of electrolyte and solvent they use). For example, I've used the TV series from Eaton recently and it's rated for operation down to -40C, and at that point its ESR is only 60% higher than at 25C. Hardly "exponential".

The average current is ~50mA, but pulsed current can touch 600mA and above. Nothing too crazy though, compared to what the GSM module's datasheet claims (2.5A peaks).

No idea why the you're observing much lower current peaks than the datasheet states. What is the max duration of the peak load?

Because of this, and board space being a consideration, I am leaning towards having a switching regulator on board to actively maintain the voltage to the GSM module. Its the only way I can think of, which will work out with board space, and without putting a truck full of tantalum caps.
Testing with an off the shelf evaluation board for a switcher (MAX18000) has given me good results so I'm optimistic. I'll use something similar I guess

Using a DCDC converter between the battery/caps and the load should increase the amount of usable energy from the capacitors, since now their voltage can droop more without affecting the load. Of course you must also make sure the battery is okay with increased voltage drop...

[BadeBhaiya]

You're looking at some fairly short duration pulses in there that require significant amounts of current. What needs to be thought about is the impedance a capacitor presents at the speeds you are looking at. If your load current pulse took 100msec to rise and fall, the supercap or bulk capacitors you have in there would be fine. It goes back to I = C * dV /dt, where the I required in that case is fairly low to keep dV/dt low. With faster pulses, you now see higher dV/dt as the I gets a lot higher and this effect is multiplied with the ESR of the capacitors. Think about it--your trace parasitics are all in the pF range, they can't hold enough charge to do anything. Your bulk capacitors have 1ohm of ESR, so if each one is delivering 250mA pulses, then they will drop 250mV each just across the ESR.

You want capacitance that has low enough ESR that you don't get these transient voltages for fast load currents and large enough that they store enough charge to hold up the output during the load current transient. I'd be willing to bet that the evaluation board you got has some reasonable MLCC capacitors on it--low ESR and overall impedance out to high frequency. You want a mix of capacitors on your board--small ones with low ESR to deal with high frequency transients (usually MLCC) and some larger size with higher ESR to deal with holding up from a bulk (DC) perspective.

Its exactly as you said, the current peaks are very quite sharp, but also short lived. The current profile is the typical GSM TX, with 577us peaks in a cycle of 4.62ms. ESR is indeed the issue, a 20F EDLC should have in theory provided more than enough current to do the transmission, but its degradation has led to this whole problem. Below is the current profile of interest, the actual transmission current.



Also, the 1ohm ESR of 220uF bulk caps is just an estimation. We won't use anything with that high resistance (or use a lot of them to counteract the individual ESR). I am honestly skeptical if its a good idea to go with a bunch of decoupling alone since it will take a fair amount to uphold the voltage, with as bad the cell is (I measured the internal resistance of the cell alone, its ~20Ω, even the degraded EDLC supercap is amazing compared to it), an active switcher (TPS61022) instead sounds like a much better option. The eval board doesn't seem to a lot of decoupling, just two tantalums I see. I suspect its high switching freq (2 MHz) is what makes it work well, so the transients are quite low.

No idea why the you're observing much lower current peaks than the datasheet states. What is the max duration of the peak load?

Its a typical GSM transmission peak. Pic is above. As for the low current reading, I'm not really sure myself. It could be that the module is transmitting at a lower power? Or the load profiler device sampling at only 100KHz (so its losing a bunch of shorter, taller peaks?). Or maybe the tiny amount of on-board decoupling is taking care of the actual peak. As I said, not entirely sure (probably low TX power, pity, I don't have control over the software)

Using a DCDC converter between the battery/caps and the load should increase the amount of usable energy from the capacitors, since now their voltage can droop more without affecting the load. Of course you must also make sure the battery is okay with increased voltage drop...

This is the route I am looking to take. The battery and the decoupling will have to deal with the higher current torture but its better than not being able to transmit. Best thing to do would be to convince bossman to let us change the battery chemistry.

[AnalogTodd]

You're looking at some fairly short duration pulses in there that require significant amounts of current. What needs to be thought about is the impedance a capacitor presents at the speeds you are looking at. If your load current pulse took 100msec to rise and fall, the supercap or bulk capacitors you have in there would be fine. It goes back to I = C * dV /dt, where the I required in that case is fairly low to keep dV/dt low. With faster pulses, you now see higher dV/dt as the I gets a lot higher and this effect is multiplied with the ESR of the capacitors. Think about it--your trace parasitics are all in the pF range, they can't hold enough charge to do anything. Your bulk capacitors have 1ohm of ESR, so if each one is delivering 250mA pulses, then they will drop 250mV each just across the ESR.

You want capacitance that has low enough ESR that you don't get these transient voltages for fast load currents and large enough that they store enough charge to hold up the output during the load current transient. I'd be willing to bet that the evaluation board you got has some reasonable MLCC capacitors on it--low ESR and overall impedance out to high frequency. You want a mix of capacitors on your board--small ones with low ESR to deal with high frequency transients (usually MLCC) and some larger size with higher ESR to deal with holding up from a bulk (DC) perspective.

Its exactly as you said, the current peaks are very quite sharp, but also short lived. The current profile is the typical GSM TX, with 577us peaks in a cycle of 4.62ms. ESR is indeed the issue, a 20F EDLC should have in theory provided more than enough current to do the transmission, but its degradation has led to this whole problem. Below is the current profile of interest, the actual transmission current.

(Attachment Link)

Also, the 1ohm ESR of 220uF bulk caps is just an estimation. We won't use anything with that high resistance (or use a lot of them to counteract the individual ESR). I am honestly skeptical if its a good idea to go with a bunch of decoupling alone since it will take a fair amount to uphold the voltage, with as bad the cell is (I measured the internal resistance of the cell alone, its ~20Ω, even the degraded EDLC supercap is amazing compared to it), an active switcher (TPS61022) instead sounds like a much better option. The eval board doesn't seem to a lot of decoupling, just two tantalums I see. I suspect its high switching freq (2 MHz) is what makes it work well, so the transients are quite low.

(Attachment Link)

No idea why the you're observing much lower current peaks than the datasheet states. What is the max duration of the peak load?

Its a typical GSM transmission peak. Pic is above. As for the low current reading, I'm not really sure myself. It could be that the module is transmitting at a lower power? Or the load profiler device sampling at only 100KHz (so its losing a bunch of shorter, taller peaks?). Or maybe the tiny amount of on-board decoupling is taking care of the actual peak. As I said, not entirely sure (probably low TX power, pity, I don't have control over the software)

Using a DCDC converter between the battery/caps and the load should increase the amount of usable energy from the capacitors, since now their voltage can droop more without affecting the load. Of course you must also make sure the battery is okay with increased voltage drop...

This is the route I am looking to take. The battery and the decoupling will have to deal with the higher current torture but its better than not being able to transmit. Best thing to do would be to convince bossman to let us change the battery chemistry.

Sounds like you're headed the right direction. The supercap degradation means you can no longer hold up the supply, so you get what appears to be a resistive characteristic every time a load step occurs.

You still may want to have low ESR caps on the line feeding your module. The higher switching frequency does not necessarily mean the loop runs faster, just that you can use smaller magnetics.

[mtwieg]

Its exactly as you said, the current peaks are very quite sharp, but also short lived. The current profile is the typical GSM TX, with 577us peaks in a cycle of 4.62ms.

Ok if the pulses are that short then other capacitor types are worth considering, like polymer tantalum. For example a couple of these in parallel. According to Kemet, each has an ESR of around 50mohm at 1kHz and 25C. At -10C, ESR rises to about 72mohm.

A properly selected supercap might still be the better option though, even if you won't be able to make use of most of its available energy. Again, hard to say without knowing more about your space constraints.

I am honestly skeptical if its a good idea to go with a bunch of decoupling alone since it will take a fair amount to uphold the voltage, with as bad the cell is (I measured the internal resistance of the cell alone, its ~20Ω, even the degraded EDLC supercap is amazing compared to it), an active switcher (TPS61022) instead sounds like a much better option. The eval board doesn't seem to a lot of decoupling, just two tantalums I see. I suspect its high switching freq (2 MHz) is what makes it work well, so the transients are quite low.

Regarding the ability to maintain output voltage during these load transients, the DCDC switching frequency is irrelevant (so long as it's not ridiculously low. Anything over 50kHz should be fine).

Its a typical GSM transmission peak. Pic is above. As for the low current reading, I'm not really sure myself. It could be that the module is transmitting at a lower power?

I'm not familiar with GSM, but I would double check that you're actually seeing your worst-case load profile.

Or the load profiler device sampling at only 100KHz (so its losing a bunch of shorter, taller peaks?).

Seems unlikely unless the "load profiler" is just poorly designed (lack of an antialiasing filter).

This is the route I am looking to take. The battery and the decoupling will have to deal with the higher current torture but its better than not being able to transmit. Best thing to do would be to convince bossman to let us change the battery chemistry.

IMO if the battery otherwise works, then changing it would be a last resort.