Power requirements: Too big for batteries, too small for mains AC
With realization of what power is actually possible to supply at my chosen cost range and level of skills I dived into DC-DC converters study. It ended with high parts count design.
Mix of 555 chips with lots of other low cost parts.
I honestly tried to find the most advanced, contemporary, easy to use up-converter and made a lot of reading about available converter chips. There is a lot of interesting, sometimes simple and sometimes sophisticated ways to generate multiples of voltage rails with integrated DC-DC converters. At the moment with all of my knowledge, there is no single chip converter which complies to my requirements.
DC-DC power supply requirements:
- The conversion must start with control inputs being unpowered
- The conversion must stop when control inputs are triggered
- After certain period, the conversion must restart no matter what control inputs do (for example software errors, poor programming, etc)
- Converter must be "stateless". There should be no latched states, requiring software resets (for example after the overloads, thermal shutdowns)
- Converter must be popular, very well known, understood to let even unexperienced users to gain usable knowledge
- There should be multiple schematics adapted by community to allow easy upgrades, contributions to schematics from community
- Converted must have zero activity (even internal generators must stay completely) shut when required
- Converter schematics must be not tied to any advanced magnetics: transformers with calibrated gaps, ferrites with more than 500 KHz requirements.
- Magnetics for converter must be easy to implement, replace, calculate. So consequently several complex to calculate or tune designs are not acceptable: inductors with high DC bias, autotransformers energy gaps and with pulse width ratios, coupled inductors designed for super high efficiency at 1 MHz, etc.
- Diodes must be a common schotky type, not a specifically rare ones
- The design must be crude, able to start and stop in a worst supply, load conditions
- Efficiency is nice to have but is not critical. Some costly parts can be avoided since users are not interested in super high efficiency.
- Efficient board space is nice to have, but not critical
- Other details: At 3-4 watt, it should feed 6 DC rails (current prototype runs at 4.85..5.15V with current 0.89..1.25A).
- Most important requirement 1: it has to output multiple voltages.
- Most important requirement 2: it must be super easy to understand for starter.
- Most important requirement 3: It must not self destruct with 0 load conditions. Who knows what users will do during the build.
At the end I finalized my choice on 555. (At least I, myself can explain every single stage and part, why it was chosen, how to debug it, how it works).
Other details about what is being changed in power supply and what lessons learned:
- The worst mistake was choosing the particular P-MOSFET. Fixed the mistake by replacing it with "logic gate level" P-MOSFET. The lesson learned is: look at actual curves in datasheets and realize that at worst gate voltages MOSFETs behave as current limiters, not as milliohm resistors. To get to milliohm range at >1A at 4V use the "logic voltage gate" MOSFET. The cost is correlated to gate voltage capability, has to pay for it, but there is good reason why it costs so.
- The PNP current limiter from Art of Electronics. It is useful only when MOSFET control is ideal - high gate to source voltage. In situation with poorly chosen MOSFET the PNP part can be removed completely with no effect on work of the circuit. I still decided to keep PNP after upgrading to ideal MOSFET. It is yet to be seen if it helps. The thought is to protect random unpredictable external supplies from overloading. The absolute limit must be 1.1..1.2A, or else the USB supplies (even ones rated for 2A) will act up. I do not want to force users to start suspecting that their USB wall-warts are at fault (even when they most likely will be at fault if device starts consuming 2A).
- Interesting limitation of 555 chip: The reset input must not exceed 555 VCC rail even few millivolts. So I had to move reset input to blue LED with guaranteed 3V level < 4.25V VCC.
- Another correction (attached schematics has to be changed): power stage with transformer (mid tap) has to be supplied from before 1N4007 diode.
- Another improvements: Add jumpers in several test points (voltage controls, for oscilloscope, multimeter probing etc), add low ohmic resistors as current test points for approximate current measurements.
- Other good find: At low power levels and relaxed efficiency requirements - copper wire of transformer coils works as a free low-ohm high power resistor. It is OK to treat it as resistor for up to 0.25W with no problems.
BTW: 1N4007 may look like another cheap offence to surrounding high cost parts. The reason to choose 1N4007 (not even lower 1N400X) is that I needed indestructible piece of very low doped silicon to have the simple isolating DC switch with leakage current at very low nA level.
Quick note: Oct 15 2015. The transformer RM10 with 12 pins is ready and has following winding details. All coils are 0.33 magnet wire.
The DC voltages after rectifiers on supercaps unloaded with input USB voltage 5.02V at the beginning of cable, 5.00..5.01V at the screw terminal.
Results: before the LM317/LM337 regulators were soldered.
5V rail unregulated: 5.19V - little bit (56mV overshoot) more than precalculated range of +4.846..5.134V
7V rail before regulator 5V analog: 7.15V is 350mV less than expected range 7.269..7.701V
15V rails before regulators: +17.42, -17.40V perfectly in range of precalculated range 16.961..17.969V
26V rail: + 27.37V perfectly in precalculated range 26.653..28.237V
3V rail: no regulator is soldered yet
One correction to the 1Hz generator duty cycle. Replacing resistor 100k with 1M, because it was mistakenly misread from earlier prototype (color bands reading mistake ?).
After this correction the duty cycle is approx. 35%, was about 9%. 350ms is charge time, 650 ms discharge time. The 7.5V rail is better now, at 7.25V.
Update: After adding LM317/LM337 regulators the switching P-MOSFET starts underperforming. It is known problem, the fix is to replace it with logic level gate P-MOSFET. Instead, before I buy a replacement part, the USB input voltage is increased 4% to be 5.18..5.20V.
The unloaded regulators inputs and outputs have following voltages:
5V rail: 6.980..7.035V input 5.032V output
3V rail: 6.980..7.035V input 3.036V output
15V rails: 17.330..17.400V inputs 13.910V and 13.960V outputs. It is much less (0.8V) than expected precalculated 14.79V.
Update October 21 2015: With replacing the FET and finding correct 1.3k resistors:
The 1.3K resistor made the rails voltages look better. Also the NDP6020P Logic Level P-FET now does not require 5.25..5.5V supply, 5.000V is enough. Resulting voltages of unloaded rails:
USB voltage +4.962V/1.36A(peak of pulse) at input of DC-DC converter
Filtered output of DC-DC converter:
-17.445V before, -15.082V after -15V regulator
+17.437V before, +14.969V after +15V regulator
+27.45V unregulated LED supply
+5.077V unregulated 5V supply
+7.103V before, +5.035V after +5V regulator, +3.039V after +3V regulator
Update October 22 2015:
After changing R54 from 0E24 to 10E:
USB voltage +4.967V/1.38A(peak of pulse) at input of DC-DC converter
Filtered output of DC-DC converter:
-17.460V before, -15.091V after -15V regulator
+17.455V before, +14.973V after +15V regulator
+27.48V unregulated LED supply
+5.116V unregulated 5V supply
+7.134V before, +5.032V after +5V regulator, +3.036V after +3V regulator
After changing R55,R56 from 0E24 to 6E8, removing J4 jumper:
USB voltage +4.965V/1.04A(peak of pulse) at input of DC-DC converter
Filtered output of DC-DC converter:
-17.338V before, -15.090V after -15V regulator, peak pulse current 32.91mA
+17.270V before, +14.974V after +15V regulator, peak pulse current 32.46mA
+27.45V unregulated LED supply, peak pulse current 29.49mA
+5.181V unregulated 5V supply, I peak 77.08mA
+7.303V before, +5.036V after +5V regulator, +3.040V after +3V regulator, I peak 77.91mA
OK. All rails voltages are in precalculated ranges now. The total USB power consumption is about 1981mW average unloaded. It is good low value, considering that most of wattage is spent on heating resistors/zeners.
So at 30% duty cycle, peak power consumption is 2909mW, with average power 872.7mW(=2909*30/100), total average USB power is 1549mW, so efficiency of unloaded (only can oscillator and 2 chips connected) DC-DC converter is 56%. The DC-DC converter itself dissipates 440mW over 9 bulky transistors and resistors and lots of diodes and zeners, making each part relatively well underloaded. Since all this parts are well scattered over 100cm^2, the power dissipation density is about 4mW per square centimeter of the board. This is was one of the goals, to avoid heatsinks and hotspots in circuit, to utilize large size of through hole parts and the board, so the assembly as a whole could be considered a single large heatsink.
Power supply input circuit.
Symbol =U= is USB Ground.
J2, L1, C1, C2 - Input filter
J4, Q1, R8, R1, R2 - Current limiter for Q4. Source: Art of Electronics 3rd Edition, p.204, fig 3.106.C.
C3, C5, R3, D1, D2 - Shoot Through protection for Q6
Q4 - Disconnector / Switch to allow noiseless measurement
Q6 - Shorting switch to ground residual noises leaking from Q4 during measurement
R4, R5 - Logic divider converting 5V to 3.3V logic signal about "Noisy vs Noiseless" state
Current limiter max current is about 1.25A. When J4 is shorted, the max current reaches 1.36A. It is designed to limit current in worst conditions, when USB source is more than 5V. The estimated limit is about 2A. This condition was never tested.