It all started when I first met ESP8266, a tiny Wifi SoC with amazing versatility and potentials.
I want to put it everywhere in my house, gathering info about temperature, humidity, presence of human (PIR), position of windows and doors, ...
But I don't want to run 100 more wires in my house. I have to rely on batteries.
Batteries have physical dimensions, have weights, and have capacity limits. And the good ones are expensive.
At the moment of writing (2015~2016), LiPo 18650 seems to have the best combination of power density (capacity/volume) and price. But, their voltage is too high at full charge (4.2v), but drop too low (2.7v) before they completely discharge.
So I need regulators.
Regulators have power consumptions, and many are quite heavy energy consumers themselves. Suppose 3400mAh @ 3.3v of energy could last an ESP 6 month, add a switching regulator with Iq (idle current use) of 2mA (actually not a very bad number, most ones you get from eBay or Amazon uses more than that), now that energy would be gone in a little over a month.
I can't imagine climbing ladders every week to change batteries somewhere in my house. That is crazy.
I need efficient power supplies, very efficient.
A simple and more efficient solution than lousy switching regulators is using a low Iq low drop-out (LDO) linear voltage regulator.
A linear regulator functions like an adaptive variable resistor -- their internal resistance are automatically adjusted according to load, so that the voltage drop on the load is always a constant VLoad, as long as the input voltage is higher than VLoad + VDrop. Due to the nature of how they work, linear regulators require very few components, and modern ones can achieve very low VDrop and Iq. For example, MCP1700 has VDrop=0.2v and Iq=2uA.
But linear regulators have two weak spots, low conversion efficiency and VIn > VOut.
-
Because regulation is done via equivalent resistance, effectively the energy carried by excess voltage is directly converted into heat. For example, when VIn=4.2v and VOut=3.3v, the efficiency of linear regulation is only 3.3/4.2 = 78.6%
-
Linear regulation cannot boost voltage, and they always have a positive VDrop, so the usable voltage range of the battery is also narrowed. For example, with VDrop=0.2v, when VIn=3.3v, VOut can only reach 3.1v.
SMPS is very different from linear regulators. They function by turning on and off the switch in the power supply circuit, at extremely high frequency, million times per second. Imagine a water dam, with water levels representing voltages. When the flood gate is open, water flow through, raising the water level of downstream; when the gate closes, water in the downstream drains and the level lowers. By carefully controlling the frequency and duty cycle of the gate, we can control the downstream water to a near constant level. In addition, by adding a pump at the flood gate, the downstream can even maintain higher water level than the dam reservoir!
Mechanism wise, SMPS is much more efficient than linear regulation, because excess energy is never discarded like in linear regulation. However, real world is never as simple. Switching on/off the gate has costs, the gate is also not completely water-tight, and there is also the cost of running the pump at the gate... all these auxiliary costs consumes the water in the dam, or the Iq of the SMPS.
Macro-power SMPS have quite high Iq, usually single or double digit mA. They are designed to handle high load current efficiently, e.g. 3~10A. When using those for micro-power applications, e.g. <1mA average current, they are grossly inefficient. Imagine a river dam with auxiliary cost about 1 ton of water each day, nobody cares about the cost when a ton of water flows in every second; but when you give the dam operator a bottle water and say: "here, regulate this." guess what look will on his/her face...
Micro-power SMPS are more dedicate versions of SMPS, which are designed to handle very small loads. Because the targeted load is small, auxiliary costs are also well-controlled. For example, TPS630252 is a 3.3v fixed voltage micro-power SMPS, with Iq of 35~82uA, hundreds of times lower than some macro-power SMPS.
However, there is still a performance gap compared with 2uA Iq of MCP1700. And this gap may become significant at extremely low load, such as 3.2uA deep sleep current of esp8266. Fully assembled ESP modules use more power during deep sleep, around 100uA. But still, there is a difference between 50% more (TPS630252) vs. negligible (MCP1700) extra power consumption.
My design aims to reduce or eliminate the Iq performance gap between micro-power SMPS and linear regulator, by combining micro-power SMPS with very efficient power supervisory circuits. The logic of my design is that, at extremely low power draw, even the supply capacity of micro-power SMPS is humongous – power generated by 0.1s of working time is probably sufficient for 1~2s of power consumption, which means that, the SMPS is idle waiting most of the time. If we can completely shut down the SMPS for those periods, we could save much more energy.
Overall, the entire circuit should only consume about 3~5uA at ultra-light load:
- TPS630252 has a shutdown mode, with current draw of only 0.1uA;
- The power supervisory chip, MC33464, has a constant current draw of around 1~2uA;
- Supporting circuit adds about 1~2uA of additional current draw.
With comparable Iq with linear regulators, much higher conversion efficiency (>90%), and ability to work when VIn < VOut, I fully expect my design to outperform any LDO solution for battery powering low power MCUs in terms of battery longevity.
This verison of circuit suffer from high voltage ripple due to the constant voltage naive shutdown control. As a result, it is not suitable for powering dedicated electronics, such as MCUs and RF components.
However, it is fairly usable for project that are not sensitive to voltage ripples, and it is very efficient!
I put this prototype in use in a PIR motion sensing nightlight, and observed surprisingly good results:
- Previously, with a bare TPS630252, the battery lasts a little over 3 months
- The new circuit has been running for over 6 months now, and there is still plenty of juice left in the battery! :)
Year 2~3 update:
- Went through a couple of battery replacement cycles, this circuit can support 8~9 months of usage on a single battery.
- The majority of savings indeed comes from the nearly complete shutdown at very low power consumption.
- For example, during the day time, when the total current draw is ~50uA, the regulator only turns on for several milliseconds every 10 seconds.
The planned improvement is to leverage low power OpAmps to factor output load into consideration.
- The power saving from the supervisory circuit is only enabled when the output load is light, e.g. <1mA;
- When the output load is high enough, simply bypass the supervisory circuit to eliminate chances of large voltage ripple.
-
V1 has a design flaw, which was discovered soon after drawing.
When Vin is higher than Vout, Vgs will be negative regardless of the !RST signal, thus renders the shutdown control ineffective.
-
V2 is also flawed, due to the failure to consider shutdown discharge circuit.
It was discovered on prototype PCB, a patch was made using diode, which confirmed the expected operation with the flaw corrected.
-
V3 has a bad layout bug, the batch of PCB is not usable.
-
V3a (same circuit, new layout) has been assembled and tested, two problems identified:
-
The circuit had to be "kick-started"
My guess is that this is a result of three factors, when combined causeing the shutdown signal to be asserted prematurely (i.e. when output reaches ~0.5v, instead of 3.3v):
- Reducing the directly attached output capacitor (to reduce energy wastage entering shutdown mode) causes low efficiency of the boost circuit at start-up -- i.e. the output voltages raises much slower
- MC33464 kind of tie !RST pin with Vin before turn-on (below 0.5v)
- TPCP8406's N-channel has very "good" response at low gate voltage
-
The output current ripple is way too large at moderate load (~2mA).
This is due to the design flaw which uses RC circuit as means to enlarge voltage monitor hysteresis -- the RC circuit delays for both asserting the shutdown, as well as de-asserting the shutdown.
A quick patch with diode across the timing resistor has been tested. The patch improves the shutdown de-assert latency, however the ripple is still fairly large.
-
-
V4 circuit is currently being designed.
PCBWeb http://www.pcbweb.com/
The circuit design and PCB layout is released under GPL license. Proprietary licensed release is available upon request.