My current quest is to choose the power regulator for an atmega328 based board which should be as efficient as possible - the supply is a (stationary) 12v gel cell battery. The whole thing should run at 3.3V and be in PWR_DOWN 95% of the time, woken up by an rtc alarm. While it is on, current drawn should typically be below ~25mA (for the atmega328, i2c rtc & barometer and rs485 chip). I plan to put the atmega328 in PWR_DOWN and wake it up with an alarm from the rtc, so current drawn when everything is shut down should be below 1mA.
Now I am wondering, should I really "bother" (as in cost, setup/layout, noise) with a buck regulator? Although power efficiency is very important for the project, the current is below mA most of the time and rather low even if on. I have been poring over loads of datasheets, and it seems to me most buck regulators that fit into my requirements draw a lot of quiescent current (e.g. LM2594: 5-10 mA's). I did find some with very low Iq (e.g. LT3991, ADP2370), but in general they have been rather expensive, difficult to source and/or are only available in awkward packages. Also it seems to me the advertised efficency doesn't really apply for my low loads, and I would be operating in the "low-power burst mode" region nearly all the time, which comes with more noise.
On the other hand, I found linear regulators with quiescent currents in the two-digit milliamps (e.g. LT1962 | LT1521 | MCP1755 | MCP1703). There are some differences in the baseline Iq wit no load and how fast Iq ramps up when load is applied, but this is the general direction I am looking at now. I have used parametric search at all the major producers, but chances are I still misseed a lot, and sometimes chips still in production are somehow not included in the tables.
Having some headroom with max input voltage and max current drawn would be nice, but I could look at e.g. over voltage protection seperately.
So apart from any other comments you might have, my main questions:
a) it looks to me as I should go with a low-Iq linear regulator. would you agree?
b) any other voltage regulators which might be a good fit, switched or linear?
c) How to calculate the power wasted by a linear regulator for a specific load - is this the correct approach?
Pwaste = (Iq * Vin) + (Vin-Vout) * Iload#with Iq at Iload
Thanks, and sorry for the lengthy post - I did try to condense it...
I understand that switching regulators are generally more efficient, but I have trouble finding a good one for my use case (see above). Can you recommend one?
I'm happy with cheap regulators from ebay, but by the sounds of it you could do with something better, thiz 12v battery how long before it sees a recharge? What's with the low energy consumption?
The battery sits at a remote station, it will get replaced/recharged ones it has run empty (although there are also some stations with solar panels). So it should run 6 months minimum, but there are other loads on the battery external to the board I am building (datalogger/telemetry, sensors).
The task of this board is to switch an external load (datalogger/telemetry unit + sensors - 12V, 30-120mA), provide time and barometer readings, measure battery voltage with ADC and communicate by rs485. Hence the low power consumption. Moreover, this will only be needed for some minutes every hour, and it will go into PWR_DOWN the rest of the time.
This is the general idea:
datalogger has power, my atmega board is active. datalogger takes measurements, sends them through gprs
datalogger has finished its task, and sends via rs485 a specific time when it needs to be turned on again
my board receives the timestamp, sets a corresponding alarm with the rtc, switches off power to datalogger + sensors, goes into PWR_DOWN
rtc alarm wakes up the atmega, power is switched to the datalogger + sensors
Apart from that routine, the board should also make available time, bat voltage and barometer readings via rs485.
I can't help but think using a small solar panel would solve all these issues... otherwise use the highest efficient dc to dc step down converter you can find... you could even look into energy harvesting...
Yeah well, that there are places that we just can not use solar panels is basically the main reason I started this. Also as I said above, the "highly efficient" step down converters I looked at provided only marginal savings over linear regulators like the MCP1703 while otherwise being a pain. I am open for any suggestions or corrections to my reasoning in the original post.
A simple solution to the high Iq of the 'decent' regulators like the MCP1703 is to use a Pch Mosfet controlled by the RTC alarm pin. By switching system power on with the RTC alarm the quiescent power can be reduced to zero. A sequence of events would have the RTC alarm switch on the mosfet long enough to power up the controller which would then assert control over the power until the reporting task is completed and then allow the mosfet to turn off.
The LM2596 has a control pin and will draw less than 100 uA in standby however the solution I mentioned in my previous post will reduce the standby to zero. The obvious drawback to using the switcher instead of the linear is the noise from the switcher however a viable solution might well be to use the switcher as a pre regulator supplying low voltage power to the linear to reduce power dissipation in the linear and control that chain with the RTC and processor controlling the aforementioned Pch mosfet to switch off the battery until the RTC alarm occurs.
How accurate is the "95% of the time in power down" figure?
If it's accurate:
Using a linear regulator, taking your 25mA figure as accurate, and ignoring the current draw when in power down mode, the average current consumption due to the load is 1.25mA.
Using a buck regulator to drop 12V to 3.3V, and assuming 80% efficiency, the current drawn from +12V is 8.6mA when operating, which is 0.43mA average at 5% duty cycle.
So when operating, the current draw of the switching regulator is about 0.8mA less than the current draw of the linear regulator, before we take the quiescent current into account. Since you can get a linear regulator with negligible quiescent current by comparison (e.g. 2uA for the MCP1703), a switching regulator is only better if its quiescent current is less then 0.8mA.
Same as the LM2594 I linked in the OP - it has a quiescent current of 5-10mA, which in the worst case would drain a 24Ah battery in 70 days without any loads attached. I can't really use the shutdown pin as I have to keep the arduino running.
Docedison:
By switching system power on with the RTC alarm the quiescent power can be reduced to zero. A sequence of events would have the RTC alarm switch on the mosfet long enough to power up the controller which would then assert control over the power until the reporting task is completed and then allow the mosfet to turn off.
That is a great idea. Currently I plan to use a rtc without battery backup (one additional part requiring maintenance, even if it should theoretically last a long time). But I will keep this in mind as a possible option.
dc42:
So when operating, the current draw of the switching regulator is about 0.8mA less than the current draw of the linear regulator, before we take the quiescent current into account. Since you can get a linear regulator with negligible quiescent current by comparison (e.g. 2uA for the MCP1703), a switching regulator is only better if its quiescent current is less then 0.8mA.
Thanks, this helps a lot when browsing datasheets. However for calculating the losses from the linear regulator, shouldn't I use:
Ilost = (Vin- Vout) * Iload / Vin
This would drop linear regulator current consuption due to the load to 0.9mA and the Iq threshold for the switching regulator at 0.4mA.
There is one more thing, looking at the datasheets for linear regulators the ground current increases with the load (e.g. LT1962, p3: 8-12mA @ 300mA). If I understand correctly, I will have to take this into account as well in the above calculation? Sadly the datasheets are sometimes not very verbose about this parameter, e.g. listing ground pin current only at zero and max loads.
I think I will use the MCP1703 for now to develop the rest of the board, and reconsider the power supply afterwards, at which point I should have more precise data about my loads, uptime etc. Still if someone happens to know a chip particular fit for this kind of situation, let me know.
oyiva:
Thanks, this helps a lot when browsing datasheets. However for calculating the losses from the linear regulator, shouldn't I use:
Ilost = (Vin- Vout) * Iload / Vin
This would drop linear regulator current consuption due to the load to 0.9mA and the Iq threshold for the switching regulator at 0.4mA.
No, the current drawn by a linear regulator from its power supply is the output current plus the ground current.
oyiva:
There is one more thing, looking at the datasheets for linear regulators the ground current increases with the load (e.g. LT1962, p3: 8-12mA @ 300mA). If I understand correctly, I will have to take this into account as well in the above calculation? Sadly the datasheets are sometimes not very verbose about this parameter, e.g. listing ground pin current only at zero and max loads.
Yes, for a linear regulator you need to add the ground current. However, for a micropower regulator, the ground current is generally very low. See Fig. 2.5 in http://ww1.microchip.com/downloads/en/DeviceDoc/22049e.pdf for typical values for the MCP1703, which indicates that ground current is typically 0.05% of load current, plus 2uA.