3 ms·

How did you design / pick the transformer? Some notes on that would help an unrelated hobby project of mine.

Animats 8y ago
Good question. It's one of the recommended ones on the LT3750 data sheet.[1] I needed to charge up 2uF to 120V in 13ms or less, and the smallest of the Coilcraft transformers listed could do that. Matching controller, transformer, and MOSFET is a big headache; subtle properties of all of them matter. That data sheet is a good read. Finding the right MOSFET was a huge headache. I kept trying reasonable ones in LTSpice, then on a real board. Gate capacitance of the MOSFET really matters. You normally think of MOSFET gates as drawing nearly zero power, but in a switcher, you want to turn them on fast, which means pumping in almost an amp for a few nanoseconds. That IC is intended for charging up photoflash units. I'm using it at a lower voltage with a lower capacitance but a faster cycle time. The examples show charge times around 1 second; I only have 22ms. I'm only charging 2uF. It's two 1uF ceramic caps; none of the current limits of electrolytic caps. Not surface mount, though; I tried those new ceramic multilayer surface mount capacitors, but they have some very strange properties; the capacitance declines with voltage. OK in filters, terrible for energy storage. It took me seven boards to get this working. The first few were a much simpler design with a 555 timer running the switcher. It worked, but it turned out I needed 2uF instead of 1uF because the Teletype selector magnet has an inductance larger than the ham community thought it did. The simpler design couldn't charge up 2uF in 22ms (one bit time); it took about 30-35ms. I had to start over with a more efficient design. This is all running off a USB port, so there's a limited power budget. Running off USB port power added of complexity. There are strict rules about drawing power from USB ports, and if you violate them, even for a microsecond, the USB port turns off, and on many devices stays off until power cycled. That's a good thing; it's why hot-plugging works. The AP2553W6 manages startup current draw and has comparators checking for abnormal situations. It has a reverse current flow detector. Spikes from the switcher made it back into the power input and tripped the AP2553W6's protection. Had to add another surface mount ferrite bead, L1, to damp out current spikes. A lot of switcher design is about putting small capacitors and inductors in the right places to damp out trouble. Most of this you can see in LTSpice. What you don't see in LTSpice is the effect of board layout. LTSpice assumes idealized wires with zero resistance, capacitance, inductance, and coupling. The layout around U1 really matters. First time around, I didn't follow the recommended layout, and the system would not oscillate. The LTSpice model is on Github, along with the KiCAD files, so you can play with the parameters. [1] http://www.analog.com/media/en/technical-documentation/data-sheets/3750fa.pdf http://www.analog.com/media/en/technical-documentation/data-...

Good question. It's one of the recommended ones on the LT3750 data sheet.[1] I needed to charge up 2uF to 120V in 13ms or less, and the smallest of the Coilcraft transformers listed could do that. Matching controller, transformer, and MOSFET is a big headache; subtle properties of all of them matter. That data sheet is a good read. Finding the right MOSFET was a huge headache. I kept trying reasonable ones in LTSpice, then on a real board. Gate capacitance of the MOSFET really matters. You normally think of MOSFET gates as drawing nearly zero power, but in a switcher, you want to turn them on fast, which means pumping in almost an amp for a few nanoseconds. That IC is intended for charging up photoflash units. I'm using it at a lower voltage with a lower capacitance but a faster cycle time. The examples show charge times around 1 second; I only have 22ms. I'm only charging 2uF. It's two 1uF ceramic caps; none of the current limits of electrolytic caps. Not surface mount, though; I tried those new ceramic multilayer surface mount capacitors, but they have some very strange properties; the capacitance declines with voltage. OK in filters, terrible for energy storage. It took me seven boards to get this working. The first few were a much simpler design with a 555 timer running the switcher. It worked, but it turned out I needed 2uF instead of 1uF because the Teletype selector magnet has an inductance larger than the ham community thought it did. The simpler design couldn't charge up 2uF in 22ms (one bit time); it took about 30-35ms. I had to start over with a more efficient design. This is all running off a USB port, so there's a limited power budget. Running off USB port power added of complexity. There are strict rules about drawing power from USB ports, and if you violate them, even for a microsecond, the USB port turns off, and on many devices stays off until power cycled. That's a good thing; it's why hot-plugging works. The AP2553W6 manages startup current draw and has comparators checking for abnormal situations. It has a reverse current flow detector. Spikes from the switcher made it back into the power input and tripped the AP2553W6's protection. Had to add another surface mount ferrite bead, L1, to damp out current spikes. A lot of switcher design is about putting small capacitors and inductors in the right places to damp out trouble. Most of this you can see in LTSpice. What you don't see in LTSpice is the effect of board layout. LTSpice assumes idealized wires with zero resistance, capacitance, inductance, and coupling. The layout around U1 really matters. First time around, I didn't follow the recommended layout, and the system would not oscillate. The LTSpice model is on Github, along with the KiCAD files, so you can play with the parameters. [1] http://www.analog.com/media/en/technical-documentation/data-sheets/3750fa.pdf http://www.analog.com/media/en/technical-documentation/data-...