Ready, set, optimize!

Alright, finally I had time to grab my solder iron and punish the naughty GSM shield for being such a greedy pig when it comes to efficiency in power consumption.

When we look more closely at the shield schematic, we can see that NETLIGHT and STATUS pins of SIM900 control the transistors which turn on and off the leds, powered from Arduinos +5V input via 300 ohm resistors R21 and R20. There's also the "shield powered" led that is always on when, as the name implies, the shield receives input power, regardless whether the GSM module is powered or not. Very wasteful when thinking about power efficiency, so I started with that first:

Actually I haven't ever done any operations on such a small SMD components before, so after unsoldering the R1 the end result was quite far from satisfactory: both copped pads under the resistor were ripped off because of too high temperature, excess force and prolonged heat exposure. Oh well, not that I need the "shield powered" LED anyway..

Before removing the R1 the power consumption of the SeeedStudio GSM shield was 12.2 mA (shield powered, but SIM900 not turned on) and 23.4 mA (SIM900 in sleep mode) when powered with external +5V lab supply (HP 6632B). After removal of R1 and thus disabling the LED, power consumption drops to 9.1 mA and 20.3 mA, respectively. This brings savings of 3.1 mA. Nice.

Wiser from this incident, I practiced desoldering a little bit with old PC motherboard and found a suitable removal procedure consisting of adding liquid flux, new solder and then removing everything with a solder wick. Then add little bit more solder to both ends of the resistor, heat one end with solder iron while lifting it with tweezers. Then repeat this with the other end and finally the resistor can be removed.

 I removed the R20 and R21 resistors and soldered a breadboard jump wire to bypass the LEDs. This time no solder pads were harmed and thus the procedure is reversible. The transistors Q3 and Q4 now connect our NETLIGHT and STATUS wires to ground when signaled by SIM900, making it posssible not only to put the GSM power and/or activity leds on the enclosure if needed, but also it's now easy to check the power status of the GSM module. Before this, the only way to see if the module is powered was to send AT commands and wait for the answer. Actually this was one of the things that annoyed me about this shield from the start. Luckily now only one digitalRead is needed.

It could have been possible to remove the transistor and tap into the NETLIGHT and STATUS pins of the SIM900 directly, but leaving them behind transistors is safer since the module uses lower voltage than rest of the device. It could have been possible to destroy the whole chip or at least the output pins with software by simply switching the pin mode on Atmega side to output and signaling +5V. Puff, there goes the magic smoke.

Since there isn't a nice way to attach strain relief to the jumper wires,  I used hot glue (again) to attach the proximal ends to the PCB to make sure all the vibration and shaking in the car doesn't cause the cables to come loose.

Now, after bypassing also the STATUS led and rewiring the NETLIGHT led (blinks once every 2800 ms, so not a concern in this case), I've managed to reduce the sleep mode consumption from 23.4 to 10.2 mA! Those LEDs really are gluttonious beasts! Then I tried inputting +12V onto the shield directly via its "low ground current, low-dropout voltage regulator", the MIC29302. Low ground current my ass. Increasing the voltage from +5V to +12V increases the quiescent current (=ground current) linearly from 10.2 mA back to 21.8 mA! And that is even without powering the GSM module! So, in this case it is better to use the 7805 to drop the voltage from +12V to +5V and distribute that to the GSM shield. If someone knows better alternatives, I'm all ears.

All in all, the ChiliCAN now consumes 14.3 mA when all the chips are sleeping (i.e., when there's no CAN traffic and no calls are being made). That constitutes maybe 90% of the time. When the heater is on (or otherwise CAN bus is active), the consumption wanders between 30 and 55 mA. During phone calls it lies around 100-200 mA having maximum peaks of 1.5 A, but those are quite short bursts, having almost nil effect on battery life. If we assume average consumption of 17.2 mA, it would take almost 4 months for ChiliCAN to deplete my 95 Ah car battery to 50% charge. I could live with those numbers :)

Soft chippy, warm chippy, little chunk of RAM. Happy chippy, sleepy chippy, purr purr purr..

Like Sheldon Cooper, also the ChiliCAN needs a good night's sleep. I've been tinkering power optimization features and I must say the results so far are pretty good!

Here's the list of tricks that I've found by reading the datasheets and various articles on the web related to Arduino power saving:

• MCP2551 CAN transceiver can be put to sleep via the Rs-pin. Just control it via one of the Atmega328 output pins via 10k resistor (+5V in sleep mode, 0V awake), and we can save around 4 milliamps.
• Atmega328p has various power saving modes, and I wanted to use the most aggressive one (SLEEP_MODE_PWR_DOWN), which reduces the power consumption to less than few hundred microamps! It can then wake up with an interrupt generated by either the CAN controller or the GSM module.
• MCP2515 CAN controller has a sleep mode as well, enabled with SPI command. Before that we must enable the CANINTE.WAKIE interrupt signaling and connect the interrupt pin to INT0/1 on Atmega328p. Then, when a new CAN frame is received, interrupt is generated and the CPU can also kicked out of his silicon bed. Warning: The first CAN frame is always discarded and does not end up into the receive buffer (I think the sleep mode of both the transceiver and controller cause this). However in my setup it doesn't concern me, since the signal from key fob light button is not in the first CAN frame in the burst of messages sent to the bus when button is pressed and the car wakes up from its slumber. Another pitfall can be found in the wake up routine of MCP2515: If we want to wake up the controller (for example when the interrupt is generated by GSM module and not the controller), then it must be done by generating the CANINTF.WAKIF interrupt by ourself, and NOT the operation mode switch by SPI interface. The latter DOES NOT WORK! This is kind of weird, since the SPI interface is specified to be active even when in sleep mode. Reading its registers seems to work, but chancing the CANCTRL has no effect. Anyway, putting the controller to sleep gives us savings of around 3-4 mA.
• Ditch the AEM and consequentially, be able to get rid of the grounding relay as well. Naturally, this has no effect on sleep mode power savings directly, but since the relay is a power hog (~ 50mA when switched on!), getting rid of it enables us to use smaller and more efficient voltage regulator (those usually have also lower maximum output current). 
• Switch from 7805 to LM2936Z-5. The 7805 has in my tests shown quiescent current (waste current generated by the regulator even without any load) of more than 4 mA, where as the LM2936 is much more efficient, having quiescent current measured only in tens or hundreds of microamps, depending on the load. Unfortunately its maximum output current is 50 mA which is way too low if we want to use the GSM module as well.  

So, putting all the chips to sleep and using other aforementioned power saving tricks, I managed to get the sleep mode consumption down to 400 MICROamps! :) Now I don't have to worry anymore about accidentally emptying the car battery!  

Now, the measurements above are done without the Seeedstudio GSM module, because it is completely another beast: 

• The SIM900 GSM chip itself doesn't consume that much in sleep, its specified to be around  1-1.5 mA, but then again it has a peak current consumption of 2 amps during network registration and other wireless activity, which makes it little bit difficult to find suitable but efficient power supply. 
• The GSM shield has its own power regulator MIC29302BU. It is a low voltage drop regulator that can use Arduinos +5V and will provide 4.1 volts (adjustable) for the rest of the GSM shield to use. It is also able to handle the peak currents required by SIM900. Unfortunately it has a quiescent current of around 8 mA when idle, which is not that good.
• The shield has a "input power on" LED as well as an actual "power on" LED that is lit when SIM900 is turned on. I'm not sure of their forward voltage drop, so it's hard to guess their current consumption. It's maybe around 2.7 mA + 9 mA when fed with input +5V, if assuming voltage drop of 2.3 V (green SMD led). Also there's a orange GSM activity led, but it's on only intermittently (one blink every 2-3 seconds).
• The total consumption of the shield when the SIM900 is not even powered  is 12.6 mA! 
• When powered and in sleep mode, the consumption increases to 23.6 mA, having bursts of 36.2 mA every 3 seconds. 
• The total consumption of the whole device (7805 as the regulator, since LM2936 doesn't output enough juice for the GPS shield) when all the chips are in sleep mode is 27.4 mA (bursts of up to 40 mA every 3 seconds). 

It's quite clear that the SeeedStudio GPRS shield v1.4 is not designed for low power consumption in mind, otherwise it would have included a more efficient power regulator and a way to either disable the LEDs or give a way to control them for example with PWM. Since the MIC29302 is intended for automotive applications and has maximum input voltage of +60V, it should  be safe to power the whole device with this regulator. However while Atmega328p and MCP2515 can operate with 4.1V, the MCP2551 CAN controller needs at least 4.5V, which is then too much for the SIM900. So, another regulator is anyway needed.  

What about LM2936 for the main board and then power the shield with car battery directly? I looked at the LED configuration in the GSM shield schematics and found out that the LEDs are powered straight from the VIN (i.e. before the regulator), and if we give the shield input voltage of +12V without changing also the resistors, LED consumption would rise to tens of milliamps (probably burning them, if not at idle, but at least when the car alternator is running..)  So in this case little bit of resoldering would be needed anyway.