Solar Freaks

Hi everybody,

I've just discovered solarfreaks.com, and as an EE student I'm fascinated with solar energy. I think it would be cool if each solar cell came with a "glue on" MPPT / regulator. Obviously I haven't thought this through much, so I'm documenting my thought process here as I go.

I think a tiny MPPT mounted on each solar cell is a possibility:

1. Lots of redundancy - no single point of failure
2. MPPT even with mismatched panels and varying light conditions (like shade)
3. Should be cheap enough to "disappear" into the cost of the panel
4. (I have some doubts - still thinking about this) Output a programmable high voltage DC, 120V / 240V / 480V - then wiring up panels and powering loads reuses existing building wiring, and any panel from any manufacturer works with any other panel
5. Grid-tie is just an automatic breaker for when the grid power goes out
6. This should also cut the cost of off-grid charger systems
7. The device could be a "smart meter" reporting on the production of the individual cell using an existing data-over-power-line protocol. Two I know of: X10 HomePNA Could probably do this "for free" since an Atmel or similar is used for the MPPT algorithm.

I'm still thinking about the high voltage DC output...

The easiest way to step up solar power to that voltage is using a boost converter. But the inductor adds a lot to the cost of the MPPT, and it needs to be high quality.

A charge pump might work better for the high voltage and low current.

The controller running the MPPT algorithm needs to consume as little power as possible, if a tiny MPPT on each cell is going to make sense. So here's my analysis of the power usage of a controller to see what is possible.

If a tiny MPPT eats up too much power generated, it doesn't make sense compared to one big MPPT for the whole solar array.

An AtTiny 12V uses 4.4 mW (2V 2.2 mA) when active, and 2 uW (2V 1 uA) when powered down.

To power it at 2V, I'd use 4 cells at 0.5V each in a string. (edit: 6 cells at 0.333V) As much as I'd like to put a controller on each cell, 0.5V is too low of a voltage.

Assume the cells are producing 1 W each, and the AtTiny 12V is using its worst-case 4.4 mW.

.0044/1/4*100 = 0.11 % lost in the controller.

Realistically, a loss more like 0.01 % is likely, since the processor will stay powered down most of the time.

Ok, so the DC-DC conversion efficiency will dominate, unless it gets up to 99.99% efficient!

Obviously, I would love to use a bigger controller and use more power, but this is a pretty good start. It looks like a charge pump is never going to be efficient enough at the high ratio I'm trying to do (2V in, 480V out, for a ratio of 1:240).

I'm getting 98.44% efficiency in simulation. The biggest source of power loss is I²R due to the transconductance of the FET.

Next I'm going to try to add a second stage at 60º phase offset - that makes 2 surface mount ceramic inductors (around 68 mH) in the first breadboard design.

Hey interesting idea indeed. I'm eagerly awaiting your results/data.

Good luck achieving max efficiency, but i think in the real world, you'll never see above 96-97% eff dealing with such low power, and high conversion ratios. Ill be happy to be proven wrong.

Also, why do you need so much voltage on the output?

TonyB wrote:Good luck achieving max efficiency, but i think in the real world, you'll never see above 96-97% eff dealing with such low power, and high conversion ratios.

I'm going to focus on the mosfet losses, but shout out any ideas if you notice I'm missing something.

TonyB wrote:Also, why do you need so much voltage on the output?

I'm going for a 120V DC output (240V and 480V later) since most household appliances run at that voltage. I agree, 2V to 120V is a big jump, but it means each of these little trackers can be hooked up in parallel and regular house wiring is usable - that'll save money.

Converting 120V DC to AC is fairly easy, and if the appliances use switched mode power supplies, they can run straight from the DC. Plus, it's less I²R in the wiring - though the Tiny MPPT might eat more power than it saves.

I'm documenting the parameters I'm tweaking to optimize the tiny MPPT.

DC in
Basically DC in is the number of cells in a string. Ideally the tiny MPPT could be glued to every cell, but that's not possible...

A 2V input is about right for extremely efficient circuits. At the lowest light levels - and at peak power, less that Voc - each cell might produce 0.333 V so 6 cells are needed. (Operating down at 0.33V is not a good solution, since that isn't enough to turn on a MOSFET: Vt = 1.5V typical, down to about 500mV.)

The max Voc for 6 cells: 6 * 0.52Vocmax = 3.12Voc

For the first build, it might be hard to get everything in 2V parts. I'll do my best on that.

Equivalent Resistance
Controlling the power produced by the solar cell is done by varying the "equivalent resistance." V=IR and all that. I'm still working on how best to control this, but here's my current approach:

Place a capacitor on the input wire. Then vary the switching voltage (Vs, below) to charge the inductor to a variable point - controlling how much energy is drawn into it to find the peak power input.

Each time the cycle starts, the inductor is pulling 0 amps. As it pulls current from the capacitor and solar cell, the capacitor voltage falls, while the voltage across the MOSFET rises. The capacitor voltage changes also, which indicates how much current went in or out of it: i(t) = C dv(t)/dt

So at time t1, measure the capacitor voltage = v1
Then at time t2, measure the capacitor voltage = v2
Assume that the solar cell produces a constant Pi during the time from t1 to t2, and that the capacitor voltage changed in a straight line (trapezoid approximation to do the integral). This simplifies the calculation to find the current the solar cell is producing:

iC = (v2-v1)/(C*(t2-t1)) - iL

By choosing t1 when the cycle starts and t2 when the switching voltage (Vs) is reached, the MPPT algorithm can calculate iC. The voltage across the inductor vL(t2) = v2-Vs. The current iL(t1) = 0, so assuming a constant di/dt (simpler to calculate):

L*(iL(t2) - iL(t1)) = v2*(t2-t1)

L*(iL(t2) - 0) = v2*(t2-t1)

iC = (v2-v1)/(C*(t2-t1)) - iL

iC = (v2-v1)/(C*(t2-t1)) - v2*(t2-t1)/L

Knowing the current produced, the power is PA = v2 * iC

Then the inductor is switched to drive the output, and only the capacitor absorbs current from the solar cell. Measure another voltage v3 at t3, when the inductor is fully discharged. (Note that v3 at t3 becomes v1 at t1 for the next cycle.)

iC = (v3-v2)/(C*(t3-t2))

And the power is PB = v3 * iC

With two pieces of data, PA (at t2) and PB (at t3), the MPPT can tell if the voltage across the capacitor needs to be increased or decreased to maximize the power. To increase the voltage across the capacitor, the MPPT can reduce the switching voltage (Vs), which will reduce the power draw of the inductor.

I wonder if there are simpler methods of measuring the current coming out of the solar cell. The capacitor voltage is easily driven up or down by controlling the switching voltage (Vs), though poor light might make the MPPT unreliable due to low voltage on all the components.

MOSFET size
The larger the MOSFET, the lower its "on resistance" (transconductance). This is the largest source of loss in the converter. 5% loss overall is considered normal, and 1% is considered amazing. I'm shooting for 1%, since that is about right for what tiny MPPT can "net" for total power production. The "on resistance" R (power loss = V²/R) is balanced by switching losses (power loss = fCV² + some V²/R).

Large MOSFETs are typically sold as "Power MOSFETs" because that's where they are used: power supplies and so forth.

Inductance
The inductance of the "flyback" controls the frequency range of the converter. A reasonable value (50 mH) is a good start. This parameter is easy to tweak at the end to optimize the frequency range so that sound coming out of the converter is above the human hearing range. Lower frequency reduces power loss during switching, so not too high a frequency (power loss = fCV² + some V²/R).

Switching voltage, Vs
At the start of the cycle the inductor conducts 0 amps - as it draws more current, the voltage across the MOSFET increases to the "switching voltage," where the MOSFET is turned off and the inductor discharges into the output. The inductor generates a high voltage when discharging, which is the "flyback" voltage boost. In other words, the switching voltage is a measurement of the current through the inductor.

The switching voltage affects a lot of things:

• Switching frequency: the inductor charges (changing voltage) proportional to input current (1mA - 40mA for a typical solar cell). However, each decision to switch occurs when the inductor reaches the switching point, so it affects how long a cycle lasts.
• Input current: the inductor draws 0 current when it starts charging. When the switching voltage is reached, it draws the maximum current, and then the MOSFET switches, sending power to the output.
• Conversion efficiency: while charging, input current flows through the inductor and into the MOSFET. Power loss in the MOSFET is I²R, so lowering the switching voltage is important for low power loss - it controls the I² in the loss formula.

Output "diode"
I used a diode in the simulator to start.

I also tested a MOSFET design (where Q1 switches the inductor to GND to charge, and Q2 switches the inductor to the output stage), and discovered that Q1 and Q2 need to be very precisely synchronized to minimize switching losses - and, if Q1 shuts off before Q2 turns on, the inductor will blow them both with high voltage.

I'll try to use the Q1+Q2 design. However, I simulated a diode design that converted 2V to 120V with 0.946% losses - the diode was peaking at 48x the current of the MOSFET (and power loss is I²R), but those peaks were very brief. A diode can be a good choice - protect the circuit with at least a 750V reverse bias voltage rating.

I think I'll stick to a single stage for my first prototype, not 2 like I said before. I'm still trying to find a better way of measuring input current from the solar cells.

I'm also looking at the competition.
Large inverters (around 5 kW systems) cost > $500 per kW.
Per-panel inverters (around 1 kW) cost $250 per kW - this is very similar to the tiny MPPT idea, but you buy a "bundle" with a panel and an inverter.

Since 6 cells can produce 11.23 W, the tiny MPPT can cost $2.808 to add up to $250 per kW.

Found a writeup of one MPPT design here: http://www.timnolan.com/index.php?page=arduino-ppt-solar-charger

See this thread about current sensing: current-sensing-t107.html#p659
I think you will need current sensing on the low side due to low input voltage.

Thanks Tony, that's a good post about current sensing.

I'm excited to start building... lots of time spent on digikey...

A good article on minimizing losses - they're doing buck, not boost, but a lot applies:

http://wind.eecs.berkeley.edu/publications/peterchev_sanders_loss_min_04.pdf

So I'll need 6 cells in series for this first build. I'd like to get as much power as possible, so something like this (link is dead) ...

I see the same panel listed lots of places, but I'm sure I'm getting burned on the price. $12.95 for 0.5Voc / 1A is $26/watt!!

So maybe what I'm looking for is a used solar panel. I'd drill into it and separate the cells into strings of 6 cells.

Searching on ebay hasn't turned up anything really exciting yet. Can anyone point me in the right direction? I'm located in the US - shipping from Australia isn't completely impossible, but I'd like to do this for as little $$ as possible...

I think I'll buy a used solar panel and modify it to suit my needs.

I'm researching price/watt on small solar panels -- mainly because the "bill of materials" for a research project like the Tiny MPPT is in the $10 range: the panel is going to be the most expensive part.

I also realize solar panel prices change rapidly, and panels for a research project can cut a lot of corners like weather protection, compatible connectors (or anything soldered at all), size, panel chemistry, and so on.

Attachments are panel prices at various times

Briefly:

1. $25 for a 2 W panel from voltaicsystems.com, but the panel is in a nice package
2. $59.67 for 63 W in individual cells from everbrightsolar.com, just the raw cells

You need to watch out when buying cells and panels like that, so they dont supply you with grade B cells.

TonyB wrote:You need to watch out when buying cells and panels like that, so they dont supply you with grade B cells.

I think you mean everbrightsolar. They have a good return policy, but of course I would carefully check each panel.

The voltaicsystems.com panels are way expensive - they had better be grade A panels

Edit: I did find some more manufacturers. Especially interesting is monocrystalline raw cells at $1.70/W. I'll edit the previous post in a bit.

Updated comparison of panel prices

Thanks dch