Finally, the timer chip and associated logic gates are working at all time intervals right across the full range.
The timer and logic circuit is shown in the photo above and occupies a footprint of 20 x 15 mm. There's plenty of opportunity for a bit more hacking, with pads for extra resistors and tracks for slight changes in the circuit, but for now it seems to work fine.
Using one of my universal prototyping boards, I rigged up some gates along with the timer chip in the hope of being able to get the full range of time intervals that should be available from the timer. So far the max has been 30 minutes before the circuit started to fail. 2 hours is the ultimate goal.
The plan is to use the basic pulse output that the timer chip seems to be able to output without too much trouble, rather than try and get the chip to 'hold' in the HIGH state before a 'release' signal is sent from the MCU. After a bit of head scratching I worked out that I could use CMOS gates to create my own hold function. I imagine that this is technically some kind of latch, I dont really know!
I firstly envisaged using an AND and an OR gate together with a feedback loop from the OR back to one of the AND inputs to create a HOLD. After looking more closely at the logic tables I realised that I needed to extend this to an XOR gate which ensures that if there are no signals whatsoever on both the inputs then the output is LOW. The theory looked good and I added a load of pull down resistors to all the inputs to stop random static from inadvertently triggering the gates.
Keen observers will note the use of a diode in the feed back loop and a momentary OFF switch to isolate the DELAY pin in the timer. when doing the initial programming. Without the diode, the circuit behaved with slightly wrong logic and I have almost no idea why! Without the momentary switch, the DELAY pin in the timer sensed resistance through the MCU and the time interval became badly affected.
So far, as tests continue, the circuit seems to be working even with all the wild array of wires flying about all over the place creating gallons of associated parasitic capacitance. Maybe this circuit is now immune to parasitics?
At last, the long awaited low power performance of the device has been tested!
I selected the TinyCurrent test module from I-Fuse for sensing the very small currents involved in this device. I liked the idea that it could be bolted right onto an oscilloscope and measure nA directly as mV. There are 3 dfferent settings on the TinyCurrent: nA, uA and mA (per mV). One word of warning though: DO NOT CONNECT ANY OTHER PROBE TO THE DEVICE CIRCUIT !!!!!! Trying to connect multiple probes causes some serious short circuiting and nasty smells.
The TinyCurrent is easy to use and I started off on the nA setting with the AEMLION uninitialised (It needs a short flash of sunlight on the solar panels to boot up). In this state, the CMOS AND gate has no voltage on either input and the total current flow for the whole device was a meager 60 nA, which equates to a discharge time of about 1900 years on a 1000 mAh lithium battery. In real life the AEMLION energy harvester cuts off the voltage to the AND gate if the battery, for some reason, discharges below 3.6V, so it's actually quite useful.
In the next test, I flashed a light onto the solar panels to initialise the AEMLION and the current shot up to a massive 240 nA, which equates to about 476 years life on the 1000 mAh battey.
Fantastic low power performance by any stretch. But what about the self discharge of the battery?
Obviously, the self discharge characteristics of the battery far out weigh any of the advantages of the the elaborate low power design and using a super capacitor is about 10 times worse. So what's the point of all this effort?
One thing to remember is that we're using an energy harvester which can pull out electrons from very low light levels and be enough to keep a 40 mAh rechargeable coin cell topped up. In this scenario, the current draw of the device is much more relevant than on, say, a 1000 mAh battery with plenty of juice from the solar panels. When the device is set up with a coin cell with a very limited light source, the low power performance of the device as a whole is more critical.
Above is the recommended layout for the TPL5110 Timer, which is done like this to reduce parasitic capacitance which in turn will mess up programming of the timer. What is does not show is that the tracks to the MCU itself should also be as short as possible.
In the current design there are some compromises as I've put in a wacking great big 6 way resistor selector switch which, even though it's right next to the timer chip, will add plenty of parasitic capacitance. Strangely, the timer chip can be still be programmed right up to 2 hours, but for some of the options, only work once! Also, the timer is fussy about how it's programmed and seems to prefer to be programmed in steps from the lowest value resistor progressively upwards. So to program a 40 minute time interval, it needs to be programmed firstly for 2 minutes, then for 10 minutes and then for 40 minutes or else it just refuses to work. 40 minutes seems to be the upper limit for the current configuration which can only be improved by getting rid of the resistor switch block and moving the timer chip and associated NPN transistor closer to the MCU. Some improvement might be achieved by changing to board material to a better dielectric specification and increasing the thickness of the board, but probably not worth the extra expense!
If I wanted longer time intervals between transmits, I'd write a 'Death Counter' to the flash memory in the ESP32 and keep shutting the MCU down immediately until the Death Counter had reach to a proscribed number. The energy wasted in booting up the MCU for one second or so should be fairly low and it's only done once every 40 minutes.
One of the problems in the previous iteration of the design was that the TPL5110 timer chip could not be programmed properly with the connection to the MCU pin used to reset it in place. This current design allows a choice of a resistor or a NC momentary switch. I tried a few resistors, but none worked so resorted to the switch, which needs to be held down when the timer and gate circuit is turned on by the slide switch. Hey presto! I can now get a time interval of exactly 123 minutes, which is almost perfect. The problem with having to wait 3 hours before re-programming the timer chip was solved by simply making a connection to GND on the slide switch so that the timer and gate power supply is not only disconnected from the battery by the slide switch, but also connected to GND when turned off.
I also decided to add a few extra features like monitoring the solar brightness and the temperature within the enclose as it could well get quite hot in full sun in the Summer. The connection from the solar source is currently a bit crude with a 1M Ohm resistor in series to a MCU pin, but it kind of works. The next iteration will have a proper bridge circuit. The DS18B20 temperature sensor worked well although a surface mount alternative would be better. Just needed to remember not to connect it to pin 2 on the MCU as otherwise the MCU cant be programmed. Pin 2 is fine for LEDs, but not much else.
Although a few problems were solved, I also managed to screw up the low power performance of the device by hacking a through hole NPN transistor into the battery voltage sensing circuit. I've currently no idea why it caused problems, but after it was removed, the battery no longer suffers from critical discharge and everything behaves as expected. It might be the type of transistor so the next PCB will have the option to include a SMT transistor or bypass it with a zero Ohm resistor. This way, if the transistor is still a problem, it can be removed without wrecking the design.
It's been moderately sunny here, so the battery got a nice boost of charge up to 4.2 volts and did not over reach the MCU pin like before. Success! Not exactly sure why it was so difficult, but believe it's something to do with internal resistors in the ESP D4 chip . Testing this device is made more difficult by the TPL5110 timer chip which tends to throw a wobbly when parts of the circuit are changed. For example, it takes a full 3 hours of being left unconnected to the battery or any other power source, before the chip becomes resetable. Could be an internal capacitor holding onto the programmed time value and refusing to let go!
Looking carefully at the photo below, there can be seen 2 resistors and a NPN transistor floating above the main board. Form the +ve battery terminal, there is the collector of the ZTX450 transistor with a 4.7K resistor soldered to the emitter. Then, there's a 100K resistor which goes to ground. The base of the transistor is activated by one of the MCU pins on one of the yellow wires and the voltage detected via the other yellow wire to another MCU pin. 3.925V gives a reading of 2923 out of a possible range of 4095 (12 bit) so there's plenty of head room for higher voltages, for example, 4.1 volts, which seems to be the maximum charge voltage on the battery.
Thankfully, after a bit of trial and error in arriving at this configuration, no diodes were required !!!
Should be easy yeah? Well, turns out it's not that easy to do this in a low quiescent power context. With a basic resistor bridge set up using 2 x 1M Ohm resistors a small amount of current is continually wasted. A rough calculation suggests that this current = 4.1V / 2,000,000 = 2 uA, but in reality current also seems to flow back through the MCU analogue read pin to earth, so the current lost could be more like 4 uA. Does not sound like a lot, but it's continuous, regardless of whether the MCU is powered on or not.
The second problem is that 1M Ohm is simply too high due to resistance on the MCU pin and means that the 3.3V threshold on the pin is exceeded. The obvious solution is to leave the 1M Ohm resistors in place but throw in a couple of diodes in series on the MCU pin thereby reducing the voltage. The technical term for this is 'forward bias voltage drop' which is an inherent characteristic of all diodes and is normally about 0.7V each, depending on the current. For a 1N4148 operating at about 4 uA the forward voltage was measured as 0.3 V, so this being the case, we'd probably want 2 or 3 of them in series between the battery and the resistor bridge.
But why not go one step further and stick a switching transistor in the mix? The transistor will also have a voltage drop across the collector and emitter, which will help improve the situation. If the transistor has low leakage, it should also be able to get rid of the continuous 4 uA current loss. So that's win, win .... hopefully.
Fortunately, the new PCBs on their way from China can be easily hacked to insert a through hole transistor into the battery measuring circuit. Should also be able to include a diode as well, if necessary. Just luck really!
There's also a graph displaying battery volts on Gadget A, as below. This gadget's quite lonely at the moment as I'm waiting for more, updated, PCBs to arrive from PCBWay. Should have chosen express delivery. Never mind.
Another thing to notice is that the actual reading for battery volts is wrong. I believe this is something to do with the ESP32 analogue read and that a look up table needs to be created for the ESP32 D4 chip. I not really too bothered about it at this stage as the new PCBs have extra high tolerance 10M chip resistors for voltage sensing and so the data will change when they start to get used. I did try putting some smaller resistors in the bridge circuit, but they would leak current back into the MCU and try to turn on some of the LEDs!
For right now, the 3.7 V battery is fully charged at 4.1 V and is out in the sunshine absorbing electrons.
The device has been outside and working now for 10 days, transmitting regularly through the day when solar power is available and through the night after a nice sunny day. However, most days it has been cloudy and even on bright days the sun just hovers on the horizon and bluntly refuses to rise any higher. It's a great test for the system as, other than on the North Pole, this is a great test environment. The AemLION did it's job well and after taking a few measurements by hand of the battery voltage every now and again, the voltage never dropped below 3.6V and so the battery was always well protected.
The next step is to get the battery voltage data onto an online database and start drawing online graphs. There will also be some tinkering with the transistor and gate circuit to try and see if there is any leakage current over the collectors and emitters. There's some new circuit boards arriving although currently stuck amongst Year of the Rabbit celebrations and COVID in China and I'm hoping to do some side by side tests using different design parameters and look at the graphical outputs for performance evaluation.
Capt. Flatus O'Flaherty ☠
The timer and logic circuit is shown in the photo above and occupies a footprint of 20 x 15 mm. There's plenty of opportunity for a bit more hacking, with pads for extra resistors and tracks for slight changes in the circuit, but for now it seems to work fine.
Using one of my universal prototyping boards, I rigged up some gates along with the timer chip in the hope of being able to get the full range of time intervals that should be available from the timer. So far the max has been 30 minutes before the circuit started to fail. 2 hours is the ultimate goal.
Above is the recommended layout for the TPL5110 Timer, which is done like this to reduce parasitic capacitance which in turn will mess up programming of the timer. What is does not show is that the tracks to the MCU itself should also be as short as possible. 
One of the problems in the previous iteration of the design was that the TPL5110 timer chip could not be programmed properly with the connection to the MCU pin used to reset it in place. This current design allows a choice of a resistor or a NC momentary switch. I tried a few resistors, but none worked so resorted to the switch, which needs to be held down when the timer and gate circuit is turned on by the slide switch. Hey presto! I can now get a time interval of exactly 123 minutes, which is almost perfect. The problem with having to wait 3 hours before re-programming the timer chip was solved by simply making a connection to GND on the slide switch so that the timer and gate power supply is not only disconnected from the battery by the slide switch, but also connected to GND when turned off.
It's been moderately sunny here, so the battery got a nice boost of charge up to 4.2 volts and did not over reach the MCU pin like before. Success! Not exactly sure why it was so difficult, but believe it's something to do with internal resistors in the ESP D4 chip . Testing this device is made more difficult by the TPL5110 timer chip which tends to throw a wobbly when parts of the circuit are changed. For example, it takes a full 3 hours of being left unconnected to the battery or any other power source, before the chip becomes resetable. Could be an internal capacitor holding onto the programmed time value and refusing to let go!
