Powering mote 24/7 using only a supercap and solar

[NeverDie]

This post is deleted!

[NeverDie]

This post is deleted!

[NeverDie]

Here's what I'm going with for now:

As you can see, this one uses only one ground plane. Board size is 22mmx18mm. It uses 0603 and 0805 discrete parts, which for me makes it easier to hand solder. The only tricky part to hand solder will be the BQ25504 itself, but there's nothing I can do about that.

[scalz]

@NeverDie
sorry for delay..
i have a pad on bottom but it's not grounded on the bottom side. the power pad is grounded only on top. i don't remember but you should have a layout guideline included with the circuit footprint, for explaining.
i also use 0603/0805.

[NeverDie]

I guess my oval pads reverted to the default pattern when I re-synced with the schematic. So, fixing that again, here's what I'm going with:


There's a bit more clean-up I should probably do, but I need to learn a bit more about Diptrace before I can do it. Meanwhile, I sent Version 6 to the fab, since fabbing is the critical path.

[NeverDie]

@scalz said:

i also use 0603/0805.

In that case, if you want to post your fab files, I can post mine. Then we can have a healthy competition to see whether they work the same or whether one works better than the other.

[scalz]

@NeverDie
oh, no competition... or let me win please I'm joking. that said i'm not sure, but I think i see Wally on your board
I'm back in an hour for sharing. i need to assemble few boards for myself

[NeverDie]

@scalz said:

i'm not sure, but I think i see Wally on your board

LOL. You probably need better ventilation when you're soldering. The fumes are getting to you.

[NeverDie]

Interestingly, in today's mail I received the "BQ25504 Stripped" PCB

https://oshpark.com/shared_projects/CYdxpE9S

that I ordered on November 16. That's about twice faster than their normal turnaround speed when I send them new fab files. The board itself is noticeably smaller than the Tindie board.

[scalz]

@NeverDie said:

@scalz said:

i'm not sure, but I think i see Wally on your board

LOL. You probably need better ventilation when you're soldering. The fumes are getting to you.

no fumes tonight, reflow oven working for me

Wally is in your GND pour, in each of your board I think.
I mean you should not have any gnd around some settings resistors and capa.
Check carefully the section 11.1 Layout Guidelines in datasheet.
This is why on the reference layout you see Top GND only on half board. Or then it may happen less performance depending of mutliple factors.
It's also explained how to route the GND.

That said, in case it can help, here is my gerbers https://www.dropbox.com/s/tugdbc3cfrlmdu7/bq25504_gerbers.rar?dl=0

Enjoy

[NeverDie]

@scalz

What is "Wally"? I haven't heard that term before, and it's not in the datasheet either. I thought you were jokingly making an obscure reference to Waldo, as in "Where's Waldo?" That's why I had thought the fumes must have gotten to you. I realize now that Wally must instead be a French electrical term of some kind.

[scalz]

@NeverDie
lol
yep you were right, where's waldo/wally but where i live we say Charlie

[NeverDie]

@scalz
Sacrébleu. I just now tried sending your gerber files to OSH PARK, but it complains, " I can't find a board outline file." A fab will need a complete set of files to get anything done.

[NeverDie]

In any event, the "stripped" BQ25504 board that I received yesterday looks rather disheveled:

@scalz
Do you see any Wally's in it?

[scalz]

@NeverDie
ahah

If it can help you, here it is : https://oshpark.com/shared_projects/XRsFBtzu
This is because i've generated the gerbers with an other fabhouse CAM.
In mine, GML is the dimension layer if i remember right.
You can also open the gerbers in a viewer like Gerbv etc..
For oshpark, i have modified it a little bit because of slightly different DRC.

[NeverDie]

@scalz
Thanks. I just now ordered it.

Interestingly, if you look at Ti's eval board for the bq25504, they do in fact connect the top and bottom ground planes, and the thermal pad on the BQ25504 is connected to the entire bottom ground plane, not just another small thermal pad:

https://www.ti.com/seclit/df/tidu094/tidu094.pdf

[scalz]

@NeverDie
cool.
Well, perhaps it's ok. I followed datasheet guidelines..no matter the powerpad is connected on top.

Perhaps, you may need the schematic to know where parts are placed

You'll need to set your resistors of course.

If really needed, no problem i can upload all the files..

[NeverDie]

@scalz

You should upload it to https://www.openhardware.io/ I'm sure a lot of people would find it very helpful.

[NeverDie]

@scalz
Out of curiosity, why did you choose such a large inductor? I notice that the "stripped" BQ25504 pcb (photo above) did the same thing. Are there advantages to the larger size?

[NeverDie]

@NeverDie said:

@scalz
Out of curiosity, why did you choose such a large inductor? I notice that the "stripped" BQ25504 pcb (photo above) did the same thing. Are there advantages to the larger size?

Nevermind. I see now from your schematic that you're using the coilcraft inductor from Table 1 of the DS.

[NeverDie]

Interestingly, looking at the Tindie board now, it's clear based on its inductor's dimensions that the Tindie board is not using one of the Table 1 recommended inductors. Perhaps that explains why the Tindie board appears to be sucking wind at the lower voltages.

[NeverDie]

@NeverDie
Coilcraft doesn't seem very available in the US, but Wurth has a similar one that's still on the DS table list that's probably a good substitute: http://www.digikey.com/product-detail/en/wurth-electronics-inc/744031220/732-1012-1-ND/1639073

Also, I'll have to update my board to use a larger inductor. I had blithely assumed, based on the Tindie board, that an 0805 would be sufficient, but now that I'm digging into it probably not.

[NeverDie]

@NeverDie
I take it back. The Wurth component I just referenced has a rather complex looking land pattern.

Fortunately, the last remaining component on the Table 1 recommended list is a Wurth that has a much easier land pattern: http://www.digikey.com/product-detail/en/wurth-electronics-inc/744025220/732-2619-1-ND/2445718

[scalz]

@NeverDie
yep from guidelines.
but also because it's better to use shielded inductors for multiple reasons like EMI..and also good inductors and parts give better perf, efficiency etc.. lot of docs on this.

That said, the footprint i use is also compatible with the 0805 advised in datasheet if i remember. it's possible to use both ref for the same footprint i've not the datasheet in front of me right now but width shoud be ok.

[NeverDie]

It seems as though TI's evaluation board ignores the design recommendations given in the datasheet, in particular: "...it is highly recommended that no ground planes be poured near the voltage setting resistors or the sample and hold capacitor." In the case of the TI's evaluation board, virtually the entire bottom layer is a ground plane. I don't know what to make of that.

Anyhow, I ordered the Wurth inductors. I also ordered a 3x3 QFN-16 solder stencil as some fallback insurance, as the BQ25504 looks like it will be very challenging to solder.

[NeverDie]

Also, the layout guidelines say, " It is best to use vias and bottom traces for connecting the inductor to its respective pins instead of the capacitors." Yet, the recommended layout doesn't do that.

[NeverDie]

Part of what's strange is the Recommended Layout (Figure 34 in the datasheet) shows pin 12 belonging to a different ground plane than pin 13. Yet, the schematic shows pins 12 and 13 wired directly together.

@scalz
Just where are the two different ground planes supposed to connect together? Or is one supposed to float relative to the other?

[scalz]

@NeverDie
ok i'll put on my git&openhardware asap.

about the planes, you're right. I've a little doubt. on most designs, it's directly to gnd. i understand about the layout guidelines. but the datasheet picture layout is a bit confusing, if it's internally connected or not..i'll check a bit later.

good point

[NeverDie]

@scalz

Section 11.1 says, "When laying out the non-power ground return paths (e.g. from resistors and CREF), it is recommended to use short traces as well, separated from the power ground traces and connected to AVSS pin 12. This avoids ground shift problems, which can occur due to superimposition of power ground current and control ground current. The PowerPad should not be used as a power ground return path."

I think that's why in the Recommended Layout (Figure 34), pins 12 and 13 are not connected directly together. However, on your board, they are directly connected, probably because of the schematic.

What do they mean by "The PowerPad"? Do they mean the thermal pad beneath the chip? Maybe that's another reason why pin 12 is not connected to the thermal pad in the Recommended Layout.

I don't know what the right answer is. I'm just trying to make sense of what seem like inconsistencies in the datasheet.

[NeverDie]

I suppose if worse came to worst, we could try answering the question empirically by using a test setup similar to this:

In this instance I partitioned the data ground from the power ground, and the two are not connected. However, via the header pins, a jumper could be connected between the two grounds. The experiment would be to try it both ways--connected vs. unconnected--and see which performs better.

[NeverDie]

The through holes on the stripped BQ25504 PCB (above)from Osh Park have too small a diameter to fit header pins into. In fact, it would require a very narrow gauge wire (narrower than what I have) to fit into it. So, unfortunately, I have doubts as to whether I will be able to test it, let alone use it.

This is all rather disappointing. So far nothing about the BQ25504 has been easy.

[NeverDie]

Not that it proves anything, but using KiCAD to look at the BQstripped PCB, I notice that the PCB designer chose to connect pins 12 and 13 both directly to each other and directly to the thermal pad.

[blacey]

@NeverDie said:

Not that it proves anything, but using KiCAD to look at the BQstripped PCB, I notice that the PCB designer chose to connect pins 12 and 13 both directly to each other and directly to the thermal pad.

Did you say KiCAD?

[NeverDie]

@blacey said:

Did you say KiCAD?

Yup. I had to use KiCAD to confirm the proper chip orientation before attempting to hand solder it to their PCB. Looks like a nice program.

I had doubts I could hand solder the BQ25504, but I tried it using Dave Jones's tack and reflow method, and to my untrained eye it looks like it came out OK, despite the chip not being completely well aligned with the solder pads:

The Wurth inductors arrived today, so I'll have a go at soldering that on next. Hopefully the BQstripped PCB designer picked the same one I did!

[NeverDie]

Looks as though the inductor hand soldered as well:

[NeverDie]

Interestingly, if one looks at the KiCad schematic for the bqstripped PCB, you will notice that it draws a distinction between regular GND and GNDPWR:
0_1480428663373_kicad_schematic.pdf

However, as already noted, that distinction seems to be cast aside when they laid out the PCB traces, since pins 12 and 13 are connected together directly by a trace. Indeed, the schematic explicitly shows GNDPWR connected to GND. So, why create the distinction in the first place?

[scalz]

@NeverDie
this is for the power gnd plane regarding the analog gnd plane. analog for instance should be connected at one point. but then in practice like you said it looks a bit different...
don't worry too much i think!
The good in this, is i think, you have learnt good things (taking care of datasheet guidelines etc...) for future

[NeverDie]

@scalz

I want to do it properly. I'm just not sure what is proper with this chip. What is the technically correct way to do it?

P.S. I ordered some 30AWG wire for hooking up the BQstripped PCB.

[NeverDie]

I had to order some 4.7uF smd 0603 capacitors and 4.42M ohm 0603 resistors.

The DS also calls for 15.62M ohm resistors, but the closest thing that Digikey carries is 15M ohm 0603 smd's, so that's what I ordered.

Meanwhile I did solder the 10nf capacitors to the BQstripped pcb, as I had those on hand. If I had it to do over, I would probably have soldered them prior to soldering the BQ25504 and the 22uH inductor, as those parts were in the way.

[NeverDie]

Just a few notes on a conservative approach to component selection for boost converters that I've lately gleaned from reading various datasheets:

1. Ideally, capacitors should be rated X5R or X7R.

2. For inductors, generally speaking the lower the DCR the better, provided that the saturation current remains appropriately high enough. For instance, the B82462G4 series (http://www.digikey.com/product-search/en/inductors-coils-chokes/fixed-inductors/196627?k= +B82462G4&k=&pkeyword= +B82462G4&pv7=2&mnonly=0&newproducts=0&ColumnSort=0&page=1&quantity=0&ptm=0&fid=0&pageSize=500 ) has much lower DCR than any of the datasheet recommended inductors. These inductors are more expensive, but their use should, in theory, lead to higher efficiency and/or lower start/operating voltages. Also, I suspect the Tindie board's performance is impaired by its choice of inductor, because I haven't seen specs on any inductors as small as theirs is that has a low DCR.

[scalz]

+1 for X5R/X7R

[NeverDie]

@scalz said:

+1 for X5R/X7R

I looked into it a little bit more, and I'm defaulting to X7R for everything, unless there's good reason not to: https://www.maximintegrated.com/en/app-notes/index.mvp/id/5527 The article also provides good reason to favor larger package sizes, up to size 1210, if the design allows.

[scalz]

@NeverDie
that's also what i do, ceramic, x7r everywhere when possible. nice article, thx for sharing

[NeverDie]

I've settled on this series of shielded inductor as having a good bang for buck ratio: low DCR, high current rating, and relatively low price: http://www.digikey.com/catalog/en/partgroup/dg-series/53965?mpart=1255AY-220M=P3&vendor=490

If anyone knows of a better series than that one, please post.

[NeverDie]

For comparison, I ordered an inexpensive ($0.91, including shipping) 6v panel: https://www.aliexpress.com/item/6V-0-6W-Solar-Power-Panel-Module-DIY-Small-Cell-Charger-For-Light-Battery-Phone-Toy/32723002222.html?spm=2114.13010608.0.0.OYll8L

It's a little bigger, but still reasonably small (a little more than 2"x3" in size). Mainly, though, the energy harvesting chips that work at very low voltages are surprisingly expensive, so maybe this higher voltage panel will allow the use of less expensive chips that have higher start voltages. Of course, indoors I don't expect it will ever get to a full 6v under load, but I do expect that under the same lighting conditions it will offer up higher voltages at the same load as the cheap garden solar cell referenced in the OP.

[NeverDie]

@scalz I was just now noticing that you like to put silkscreen between component pads:

Does silkscreen work the same as solder mask, in that it sorta "repels" molten solder away from it? I really don't know, but I've been assuming that it doesn't. In fact, I've gone out of my way to move silkscreen away from the solder pads out of fear it might contribute to inadvertent solder bridging between pads.

Anyone know?

[scalz]

@NeverDie
well, sometimes i think to not put them at all! especially when i want to make things compact..at a moment, it becomes difficult to fit them, and you can't reduce fontsize because it's ugly. usually, i use 32mils.

I've never had short because of this though.

[NeverDie]

@scalz said:

@NeverDie
well, sometimes i think to not put them at all! especially when i want to make things compact..at a moment, it becomes difficult to fit them, and you can't reduce fontsize because it's ugly. usually, i use 32mils.

I've never had short because of this though.

Actually, I wasn't referring so much to the letters and numbers--which are off to the side-- so much as the little dashes and lines that appear between the pads. The letters and numbers are useful.

[scalz]

I know, you're right, these are useful
but when doing compact stuff, it's not always easy to display everything well..
I usually have my computer display in front of me when assembling, or i print the layout x4..so i don't look at the silkscreen so much..

About "repels", if i understand right, i have never got short during soldering. I think you mean silkscreen, between parts, would help to bridge soldering? but no never got problem.
I often use stencil+reflow oven, so there is only the solder needed. For proto, generally i almost assemble at hand, still no problem

[NeverDie]

In that case, maybe it's a non-issue, which would be good to know for future reference.

[NeverDie]

@scalz

I see now why your board has 10 resistors instead of TI's reference schmatic's 9 resistors. It's because there doesn't seem to exist an off-the-shelf 15.62M-Ohm 0609 resistor, even though TI apparently thinks such a component does exist. I know Digikey doesn't stock any 15.62M-Ohm 0609 resistors, that's for sure. So, your two series resistors are probably a 15M-Ohm resistor and a 620K-Ohm resistor, both of which Digikey sells.

[NeverDie]

@scalz
I worked out the following as the mapping from resistor labels on your board to resistor values (assuming the target is Figure 14 from the DS):
R1 = 4.42M
R2 = 15M
R3 = 620K
R4 = 1.43M
R5 = 4.22M
R6 = 4.42M
R7 = 4.02M
R8 = 5.9M
R9 = 4.42M
R10 = 5.6M

[scalz]

i don't remember, i'll look at this later. but i posted the schematic some posts above if it can help.

[NeverDie]

From the fab I've already received my board, your board, and obvioiusly the "stripped" board (photos above). On Thursday I should receive from Digikey resistors with precisely the values in the above list. I'll then be able to finish assembling the boards and see how their performance compares. TI has a paper for measuring the BQ25504 efficiency: http://www.ti.com/lit/an/slua691/slua691.pdf

[NeverDie]

@scalz
I assembled both your board and my board today and did a quick test. The good news is that both boards appear to function with a cold start voltage of about 330mv, and for that reason it appears that both boards outperform the Tindie board. I am guessing that is, at least in part, because they use one of the datasheets recommended inductors, as compared to the Tindie Board, which does not appear to.

[NeverDie]

I've played with the BQ25504 boards only a little bit thus far, but the truth that the datasheet glossed over in the executive summary section is that at the very low voltages (330mv+), the efficiency is terrible up until it charges up the caps to reach a threshhold voltage of around 2 volts or so. Those ultra-low voltages are the "cold start" phase, and the aim of that is, apparently, to bootstrap itself using available energy up to the threshhold voltage (maybe there's a better term for it). So, the trick to using this chip appears to be to maintain at least the threshhold voltage on the caps. You really don't want to ever invoke the cold start mechanism, because the the cold start efficiency appears to be quite miserable. So, while it is true to say that it works at ultra low voltages, it's just a partial truth that leaves a lot unsaid.

So, if one starts with the assumption that a relatively high voltage (say 2 volts, give or take) must be maintained on the caps in order for this to be practical, it really has me wondering now how this chip compares to other boost chips under the same assumptions. Fortunately, I have test boards for a number of different kinds of boost chips already in the pipeline.

[NeverDie]

A closer reading of the cold start section of the datasheet (Section 8.4) confirms my initial observations. VSTOR_CHGEN is what I was calling the "threshhold voltage" and is typically around 1.8v. VSTOR must be greater than VSTOR_CHGEN before cold start ends. The datasheet conceeds that cold start runs an "inefficient" boost converter. Just how inefficient? Well, I haven't seen any curves in the datasheet that answer that precisely (perhaps they were deleted by the marketing department?). Apparently, if you want to know that vital information, you have to compile your own data. I guess TI either doesn't know themselves (unlikely), or they're just not sharing.. Gee, thanks TI

However, the reality is worse than what I initially surmised, especially if using a supercap: That's because a discharged supercap will crash VSTOR, and so the chip could remain stranded in cold-start mode. Quoting the datasheet: "When large, discharged super capacitors with high DC leakage currents are attached, the intial charge time can be signficant." Yeah, no kidding. But, it appears there's also a notion for how to workaround this problem: "It is highly recommended to add an external PFET between the system load and VSTOR. An inverted VBAT_OK signal can be used to drive the gate of this system-isolating, external PFET."

So..... now I need to figure out the brass tacks of how to do that and probably redo the PCB to include such a PFET.

[NeverDie]

I think I'll try using a load switch. That way I can use the BAT_OK signal directly, without having to invert it. What I don't presently know is how fast the BAT_OK signal changes. If it lags considerably, then there's a risk that VSTOR might fall back into the cold start region before the current draining from vstor into the supercap can be shut off.

[NeverDie]

I found this curve:

not in the datasheet, but here: http://www.ti.com/lit/ml/slpt026a/slpt026a.pdf

You'll note that it shows VSTOR = 3.0V, which I think is probably key to getting anything worthwhile from this chip. Anyhow, as long as VSTOR=3.0V, it's a pretty impressive curve.

[NeverDie]

TI makes an "ultra low leakage load switch" which might be a good fit: http://www.ti.com/lit/ds/symlink/tps22860.pdf

[NeverDie]

OK, I just put together a breakout board for the TPS22860 to audition it for this new role. I just now sent it to the fab, so it will be another two weeks (!) before I'll know. I just hate how these fab iterations stretch these projects out far too long.

[NeverDie]

So, what should the programmed voltage cut-off's be? Here's my first pass at that:

VBAT_UV = 0 volts. // it's a capacitor, so you can't overdischarge it.

VBAT_OV = 3.6 volts // because we don't want to run the mote at more than 3.6 volts.

VBAT_UV_HYST = 1.9v. // Note sure. To avoid the cold start inefficiency if the supercap is flat, it shouldn't be lower than 1.8v. If the VSTOR cap is ultra low leakage, then maybe 1.9v would keep the BQ25504 out of the cold start zone for long enough (?). If not, I might need to utilize a higher capacitance VSTOR.

VBAT_OV_HYST = 2.0v. // Another guess. Needs to be higher than VBAT_UV_HYST, or there won't be any charge to transfer to the supercap. On the other hand, a higher voltage would lead to greater inefficiency if the supercap voltage is low.

As long as the supercap's voltage is above VBAT_UV_HYST, then the supercap and VSTOR will remain connected.

[NeverDie]

Translating the above programmed voltages into resistor values that can be purchased on Digikey yields the following:

ROV1 = 5.1M
ROV2 = 4.7M
ROK1 = 6.2M
ROK2 = 3.3M
ROK3 = 510K
RUV1 = 10M
RUV2 = 0 or 1 ohm (doesn't matter which)

[NeverDie]

I received the MCP1640 boost converter board from the fab yesterday, and so I put it together. It works, but I can see now why it' (and probably most boost converter boards not intended for solar) is a bad fit for solar: basically the lower the input voltage, the higher the input current that's required in order to generate current at the output voltage. However, under low light circumstances, when a small solar panel will have low voltage, it is precisely the time when it also won't be able to supply much current either. So, without modification, it's a poor match for what's needed.

What would be needed to make it work is a mechanism to throttle the output voltage and output current to better match the limited input voltage and input current that's available. I suspect that mechanism is basically what MPPT is. So, if one were to replace the fixed resistors on the MCP1640 board with a potentiometer, one could manually do MPPT. But that obviously isn't practical, so one could instead use an mcu and a digital potentiometer, together with voltage and current sensors, to accomplish the same. That would probably work, but it leaves open the bootstrap problem: how could you power that equipment from a cold start scenario?

So, I think this does give insight as to why the BQ25504 behaves as it does. Rather than using an mcu and generic voltage and current sensors, it uses specialized hardware to accomplish the same result. However, just like an mcu, that hardware can't work from a cold start, and so the whole setup wallows in high inefficiency until it can emerge from the cold start.

[NeverDie]

On the other hand, since I'm sure to have an mcu in just about any mysensor-type project, why not leverage it? If there exist (?) low-voltage boost converters that are easily mcu controllable, that might be worth exploring.

[mfalkvidd]

@NeverDie you have probably already seen this, but in case you havent: mouser has a great article on mppt http://www.mouser.se/applications/solar-panel-power-tracking/

http://www.limpkin.fr/index.php?post/2011/12/07/Indoor-solar-energy-harvesting%3A-a-platform-to-(finally)-get-some-numbers has some hard data on how much power one person managed to get.

I haven't built anything like this yet, but I am following your progress because I would like some of my sensors to be self-powered in the future.

[scalz]

i read this one too. his next article is http://www.limpkin.fr/index.php?post/2012/03/20/Indoor-solar-energy-harvesting%3A-3-months-data
and i studied this one http://www.ti.com.cn/cn/lit/ug/tidub22b/tidub22b.pdf when i did my choice about overall cost, if i wanted reliable delivery and signing packet, for my multisensors..so i thought that would add too much overhead on my specific device (cost) but this chip is very nice for simple, low power BLE like, short burst

[NeverDie]

Thanks for the articles, guys.

The TI article takes the approach of running off a battery when there's not enough solar. Although that feels a bit like cheating, it might be valid if used just to power an MCU rather than do a full cold start before that's possible. Another approach would be to use something like an LTC3108, which is a boost converter optimized to boost very, very low voltage/current values. Maybe that could power an MCU much more quickly than rely only on the cold start abilities of the BQ25504. Perhaps it would turn out that an LTC3108 would be sufficient, all by itself. I think it's worth exploring.

[NeverDie]

Also, a larger, higher voltage panel coupled with a buck-boost converter (https://www.openhardware.io/view/276/33v-Buck-Boost-DC-DC-Converter) might offer another solution, perhaps one that isn't so affected by cold start issues.

Of course, some might simply use a larger, higher voltage panel directly wired to the supercap, without any intermediating converter. I'm not sure how that compares in performance, but it does have the virtue of simplicity.

[mfalkvidd]

Jeelabs has some nice posts on startup power:
http://jeelabs.org/2013/03/29/jeenode-micro-start-up-power/index.html
http://jeelabs.org/2012/12/29/rfm12b-startup-power-consumption/index.html
http://jeelabs.org/2012/09/09/delayed-power-up/index.html
http://jeelabs.org/2012/05/29/its-about-survival/index.html

[NeverDie]

I built the pass-through boost converter (https://www.openhardware.io/view/285), and it works. However, for use with a small, low voltage solar cell, it has a cold-start problem: in order to enable the boost converter, the SHDN pin must be held high by continuously applying at least 1 volt to it.

[mfalkvidd]

@NeverDie would the mosfet solution suggested in the delayed power-up post from jeelabs work?

[NeverDie]

@mfalkvidd
I guess you're seeing something that I don't, because I don't see how it relates.

[mfalkvidd]

@NeverDie no, I probably know too litte to see that it is not applicable

[NeverDie]

@mfalkvidd
Aw gee, I was hoping you saw something in it that was blindingly obvious that I was, uh, blind to. In fact, I'm still hoping that.

jcw is a really good author. Too bad that lately he has gone off the deep end into FORTH and the like rather than exploring solar energy like he should be. LOL. Then maybe I wouldn't have to be doing it. Surely someone out there has figured out this topic already. I mean, solar calculators have been around for decades, and they have tiny solar cells. Maybe I should use one of those solar cells? Now it has me wondering what voltage solar calculator cells produce.

[NeverDie]

Moving on, I built a different boost converter using the LTC3525-3.3 (not the LTC3525D-3.3, as previously, which was the bypass version of the chip).

The LTC3525-3.3 is, I think, a better choice than the D version, because when the voltage on SHDN falls below 0.4v, it disconnects itself from the output. That means the supercap won't drain itself to zero by draining current backwards through the boost converter. When the voltage on SHDN is greater than 1.0v, then the chip is reconnected to the output (a supercap in our case). What happens when the voltage on SHDN is between 0.4v and 1.0v? Good question. That's undefined.

[NeverDie]

I put together my buck-boost project (https://www.openhardware.io/view/276/33v-Buck-Boost-DC-DC-Converter), and it works. I thought that if I were to charge a supercap (or series of supercaps) to a higher voltage (say 5v), it might be a good way to use the stored energy. At the moment, though, I don't think that's the near-term direction this solar project is heading.

[NeverDie]

@NeverDie

I assembled and tested the cheapest of the boost converters (https://www.openhardware.io/view/278). It works. Ironically, because it lacks an enable pin, one could argue that it actually has less of a cold start problem than its more expensive counterparts. Its startup voltage is 0.8v, and after that it can drop to 0.5v (where it draws about 12ma of current in a no load scenario). For weak indoor lighting, its current demands are too high for a small solar cell. However, in direct sunlight, it could definitely provide 3.3v.

[NeverDie]

Good news! I jerry rigged one of the above mentioned load switches inline with the BQ25504, and Lo! It can sustain a lightly loaded voltage of 3.1v output all the way down to an input voltage of 0.21v before collapsing.

So, I hooked up my el cheapo solar cell, and I confirmed that it does the business even in ambient light.

[NeverDie]

So, I've revised my BQ25504 PCB to include the load switch, and I've sent the files off to the fab.

[NeverDie]

The next thing I need to look into is how to measure voltages and currents in the least disruptive way. Using a multimeter to measure the voltage on a cap can noticeably affect the voltage, due to current drain. One option would be to use the ADC on an atmega328p. Another would be to use an INA219. I think one of the keys will be to switch on for the measurement and then switch off as quickly as possible, as opposed to a multimeter which stays switched on until you manually switch it off.

If anyone has any other suggestions for minimally disruptive measurements, please do post.

[NeverDie]

@NeverDie said:

Translating the above programmed voltages into resistor values that can be purchased on Digikey yields the following:

ROV1 = 5.1M
ROV2 = 4.7M
ROK1 = 6.2M
ROK2 = 3.3M
ROK3 = 510K
RUV1 = 10M
RUV2 = 0 or 1 ohm (doesn't matter which)

I built a bq25504 pcb using these values, and it tested out correctly.

[NeverDie]

I found that this guy had a similar goal as me (in his words, "I envisioned several electronic projects that would charge a capacitor during daylight using a solar panel, and then consume that energy to stay awake at night") and has done an interesting write-up not only on how to do non-disruptive capacitor voltage measurements, but also how to reduce capacitor self-discharge (e.g. even "no clean" flux residue can significantly increase self-discharge of capacitors and should be removed). To make non-disruptive voltage measurements for monitoring/debugging of prototypes, he uses Microchip MCP6S22 to create a high impedance buffer with 10 trillion ohms of input impedance.

[NeverDie]

This article gives a bit more detail on the brass tacks of how to actually use the mcp6s26: http://hackaday.com/2009/03/30/parts-programmable-gain-amplifier-mcp6s26/

I just now ordered some mcp6s26 from Digikey so I can try them out.

[NeverDie]

@NeverDie said:

I suppose if worse came to worst, we could try answering the question empirically by using a test setup similar to this:

In this instance I partitioned the data ground from the power ground, and the two are not connected. However, via the header pins, a jumper could be connected between the two grounds. The experiment would be to try it both ways--connected vs. unconnected--and see which performs better.

By the way, I built this, and it unambiguously confirms that GND1 and GND2 need to be connected in order for the BQ25504 to work. If GND2 is left floating, it doesn't work.

[NeverDie]

I finally received the PCB breakout I made for the load switch. I put that together. Then I soldered together a prototype board to probe with an oscilliscope to see what's really going on with the BQ25504:

I soldered in a 300uF ceramic capacitor across VBAT and GND as the "battery". The datasheet recommends either a rechargeable battery or a minimum of 100uF.

The first scope shot is with the 300uF nearly depleted. The blueline is the voltage across VBAT (i.e. the 300uF capacitor in this particular setup). The yellow line is the voltage on VSTOR on the BQ25504 breakout (about 4.7uF):

What you see is VSTOR being charged by the solar cell up to around 2 volts and then very briefly setting VBAT_OK to high, which turns on the load switch to discharge VSTOR into VBAT until VSTOR drops down to around 1 volt. Then it repeats. Notice that the voltage across VBAT doesn't climb very much in this screen shot. That's because we're in the cold start phase, and the conversion isn't MPPT but rather something else that's highly inefficient.

Note: the amount of light landing on the solar cell is exactly the same across all the scope shots, because it's just a ceiling light in the room where I'm taking the measurements. Keep that in mind when you compare this first screen shot to the last one.

[NeverDie]

Later you can see that the voltage on VBAT has risen to 1 volt, and pretty much the same thing is going on:

Later still, the voltage on VBAT is about 1.6 volts, and again it's the same sort of thing. Notice that the voltage across VSTOR drops all the way to the voltage across VBAT, so as the voltage across VBAT increases, the period of the cycles becomes shorter:

This next scope shot shows that as the voltage on VBAT rises to around 1.8 or 1.9v, a transition happens, and VBAT_OK stays turned on permanently. Notice how the rise in voltage after that really starts to accelerate. That's because at that transition, MPPT finally kicks in:

Lastly, here's anothe scope shot showing that after MPPT kicks in, the voltage rises a full volt in the same time period where previously (see preceding scope shots) hardly any voltage increase across VBAT was measured:

All of the above proves quantiatively how important it is to keep the voltage on VBAT above 1.8 volts, so that the BQ25504 chip can do an efficient job of harvesting the energy. As I mentioned earlier, the amount of light falling on the solar cell is exactly the same in all of the scope shots.

[NeverDie]

Here's a final screen shot which documents something different, but no less important:

In this example , the yellow line is the same as before (voltage on VSTOR), and it's the scenario where the transition from MPPT off to MPPT on occurs. However, in this case the blue line is the voltage on VBAT_OK. Thus, you will notice that each time VBAT_OK goes high (albeit very briefly), VSTOR discharges itself into VBAT. It does that because I wired VBAT_OK to the ON/OFF pin of the load switch. Thus, when VBAT_OK is asserted, the load switch closes and turns ON. When VBAT_OK is not asserted, the load switch opens and turns OFF.

I think "VBAT_OK" was horribly named given that it's not just a flag for that, but, as shown, the one and only way that the BQ25504 triggers the load switch to discharge each accumulated charge on VSTOR into VBAT. Honestly, without a load switch, or equivalent PFET circuit, I'm not sure how well, or even if, the BQ25504 would work. TI does recommend it, but the datasheet buries that recommendation pretty deeply. It's not even shown in any of the DS's application schematics. I'm not sure why that is, because, as illustrated by this scope shot, it seems absolutely essential. Indeed, that is why I revised my BQ25504 breakout PCB to include one.

[NeverDie]

Now that I've got a solar charger that works, the next step is to find some kind of capacitor that can hold its charge reasonably well through at least the night, and preferably a lot longer. Rather than jump immediately to supercaps, I thought I'd first try just a regular ceramic capacitor, as I've heard those hold their charge pretty well.

So, I soldered a 330uF ceramic SMD capacitor to a header:

I put a charge on the capacitor, and then I connected it directly to the leads of an oscilliscope:

Here's the scope scrape from over a few minutes:

As you can see, the voltage on the capacitor is dropping fairly rapidly! But why? Does the capacitor have a high self discharge? Is the header plastic too conductive? Is the impedance on the oscilloscope too low? All of the above? None of the above? Something else?

[NeverDie]

To address the plastic conductance question, I soldered leads directly to the cap:

and re-ran the test:

Because the starting charges were different, it's not a complete apples-to-apples comparison. Nonetheless, it's still clearly bad even without the plastic.

[NeverDie]

So, I decided to solder the 330uF cap to an mcp6s26, (see earlier post above for a description) which I had just recently received from Digikey:

to see how that compares:

What's evident is that the rate of decline is less, and it seems to improve with time.

There is some decline still evident. How much of that is due to the capacitor and how much is do to ongoing measurement using the mcp6s26?

[NeverDie]

I kept the test running after the last post, and now, a bit more than four hours later, the scope shows:

So, that's reason for optimism. I have to cut the measurements short tonight, but I'll try again tomorrow morning, when it can have more run time.

[mfalkvidd]

@NeverDie Just a suggestion: keep the cap on charge during the night, since it seems they get better if they get tie to saturate.

[NeverDie]

Thanks for the suggestion. I decided to keep the above setup running through the night after all, and now, about 5 hours after the last screen shot, it's showing a voltage of about 2.8 volts. I'll let it continue to run and see how it goes.

Meanwhile, I ordered a shotgun assortment of different caps from Digikey. Unlike capacitors from Aliexpress, they'll have reputable datasheets. They'll have differing capacitance levels (220uF all the way up to 90F) so I can start to get a bearing on how much capacitance might be needed. Ideally I'd like to find something that could hold enough charge for a week or more, to handle worst case scenarios, but as yet I'm not sure that will be possible. For instance, if it turns out that 90F is what's needed, I'm not as yet sure if my el cheapo solar cell powered by ambient light could even charge it in a day. On the other hand, I'm quite sure it could easily charge a 330uF capacitor. So, with the new caps, I'll start gathering some data on how much charge can be realistically harvested as well.

The 220uF cap (http://www.digikey.sk/product-detail/en/NOJC227M006RWJ/478-8864-6-ND/4562183) will be interesting, because it's made from Niobium Oxide, which has low ESR and claims to have a leakage current of 14ua. Leakage current and self discharge rate are defined differently, but my hunch (?) is that they are strongly correlated (i.e. my WAG is that a cap with a low leakage rate will have a low self discharge rate also). That is to say: I'm hoping that leakage current might be a proxy for self discharge rate, because at least so far it has been easier to find info on leakage rates than self discharge rates.

[NeverDie]

@mfalkvidd said:

@NeverDie Just a suggestion: keep the cap on charge during the night, since it seems they get better if they get tie to saturate.

You're right! According to Maxell (http://www.maxwell.com/images/documents/1007239-EN_test_procedures_technote.pdf), the way leakage rates are measured is to first hold the capacitor at its rated voltage for 72 hours. For measuring self discharge, the cap is held at its rated voltage for just one hour, but then it is measured over the subsequent 72 hours.

On StackExchange, somebody asked, "How to calculate self-discharge time of capacitors given the leakage current?" (http://electronics.stackexchange.com/questions/35568/how-to-calculate-self-discharge-time-of-capacitors-given-the-leakage-current). There was only one answer, and its bottom line was: "the only way to be half-way sure about the self-discharge rates will be to build up a bunch of prototypes and test them." I'm surprised there's not a better way, but none was posted.

[mfalkvidd]

@NeverDie the robot room link you provided earlier (to be precise, this post and this post) has more details, especially this part:
because there is no standard for testing self-discharge (how long to charge, when to start measuring current), it is difficult to compare different capacitors to determine which really has the lowest leakage for your project.

[NeverDie]

The Maxell paper contradicts him in that regard. It says, "The test methods for leakage current and self-discharge are consistent industry wide. " That isn't to say, though, that the information for a particular capacitor is easy to find, however.

The capacitor that I've been testing is this one: https://www.digikey.com/product-detail/en/murata-electronics-north-america/GRM32ER60G337ME05L/490-13976-1-ND/6155806 It has been about 15 hours now since I started the test, and the voltage is presently at about 2.6 volts. So, as a first attempt, it's encouraging to see the capacitor maintaining a worthwhile voltage through the night until the next day.

[NeverDie]

By the way, it's worth mentioning that the mcp6s26 defaults to a gain of 1x on Channel 0. In my case I'm using exactly those defaults, which is convenient because I get the chip's benefits without needing to bother with SPI programming of the mcp6s26.

[NeverDie]

Maxim Integrated has a handy online calculator for estimating the super-capacitor capacity needed for RTC backups: https://www.maximintegrated.com/en/design/tools/calculators/product-design/supercap.cfm

I don't believe it accounts for self discharge, but maybe it at least helps estimate the right order of magnitude.

It has been close to 24 hours now, and the the voltage on the above test capacitor is now at 2.5 volts.

So, I think the next step is to hook one up to a sensor mote and see how long it can go--having the mote measure and report he capacitor voltage say every 5 minutes or so--before running out of juice.

[NeverDie]

I finally found some proper supercapacitor discharge curves (from page 15 of http://www.nec-tokin.com/english/product/pdf_dl/supercap_manual.pdf

So, based on that, it looks like the FY Series of the NEC-Tokin supercaps is worth looking into. According to the chart, the best performing one is this one: http://www.digikey.com/scripts/DkSearch/dksus.dll?Detail&itemSeq=215649477&uq=636192387876420325

They sound well adapted to supplying voltage to a sleeping mote. Whether they can handle the current demands of an RFM69HW at maximum Tx, at least for a short burst, is unclear. If not, then perhaps it could be used to charge up a more conventional cap in advance of the burst current being needed.