Post by thread:[EE]: SMPSU design: The Story So Far....

OK folks, I've spent the past week meddling with my battery charger design, and I've come to the following conclusions: - The Atmel AVR450 appnote is just plain wrong. The section on charge profiles and charging methods is OK (but check against the Sanyo and Panasonic charging specs before you go writing any code). - Apparently the AVR450 drive circuit doesn't turn the MOSFET on fast enough, so it ends up cooking (there are a few people on the AVRFreaks forums saying this). The circuit is a 680R from gate to source, a BC847 (SMD BC547) pulling the gate down (E to GND, C to FET gate). The 847 gets its base drive from the MCU via a 1k series resistor, and a 10k pulldown on the base side of the 1k. Power goes into the source, drain is wired to the inductor/diode junction. If my description isn't good enough, google for "AVR450 pdf" and take a look at page 27. - Assuming it is slow switching that's cooking the FET, how could I speed up the switching circuit? Some of the messages on AVRFreaks suggest an LM2725 high side driver, but they're obsolete and the replacement (LM27222) is "just a bit" expensive (plus it's designed for push-pull MOSFET drivers, not single-ended). - I was going to drive the gate of the FET directly from the PIC (and maybe use an N-channel instead of a P-channel FET), but I'm not sure if this would work. It certainly doesn't work in SPICE, but since when have SPICE simulations ever accurately modelled real life? :) - Microchip AN793 seems to be pretty accurate. The formulae on <http://www.smps.us/> (buck converter section) seem to agree with the appnote, and a lot of the other stuff I've looked at says the same thing. - I've come up with an inductor value of 68uH for this scenario: Vin = 6V, Vout = 1.5V, Iout(max) = 1.5A, Iout(min) = 10uA, Fosc=100kHz, Vripple [permissible ripple] = 10% of 1.5V = 15mV - Does this seem about right? - If I pick a maximum Vout of 3V (i.e. two NiMH cells), I get L=75uH. If I want to optimise my charger so it charges one or two cells at a reasonable efficiency (70% or better), should I stick to the lower value, or pick one in the middle (if one is available)? Or maybe pick the higher value? Common sense suggests "pick the one in the middle" but I'm not sure how things are generally done (a lot of the stuff I've found is a bit hazy on this). Thanks. -- Phil. | Kitsune: Acorn RiscPC SA202 64M+6G VF+UniPod spam_OUTphilpemTakeThisOuTdsl.pipex.com | Cheetah: Athlon64 3200+ A8VDeluxeV2 1G+180G http://www.philpem.me.uk/ | Tiger: Toshiba SatPro4600 Celeron700 256M+40G

Vripple [permissible ripple] = 10% of 1.5V = 15mV 10% 0f 1.5V is 150mV not 15mV, probably just a typo :) Kevin On Fri, 26 May 2006, Philip Pemberton wrote: {Quote hidden}> OK folks, I've spent the past week meddling with my battery charger design, > and I've come to the following conclusions: > > - The Atmel AVR450 appnote is just plain wrong. The section on charge > profiles and charging methods is OK (but check against the Sanyo and > Panasonic charging specs before you go writing any code). > > - Apparently the AVR450 drive circuit doesn't turn the MOSFET on fast > enough, so it ends up cooking (there are a few people on the AVRFreaks forums > saying this). The circuit is a 680R from gate to source, a BC847 (SMD BC547) > pulling the gate down (E to GND, C to FET gate). The 847 gets its base drive > from the MCU via a 1k series resistor, and a 10k pulldown on the base side of > the 1k. Power goes into the source, drain is wired to the inductor/diode > junction. If my description isn't good enough, google for "AVR450 pdf" and > take a look at page 27. > - Assuming it is slow switching that's cooking the FET, how could I speed > up the switching circuit? Some of the messages on AVRFreaks suggest an LM2725 > high side driver, but they're obsolete and the replacement (LM27222) is > "just a bit" expensive (plus it's designed for push-pull MOSFET drivers, not > single-ended). > - I was going to drive the gate of the FET directly from the PIC (and > maybe use an N-channel instead of a P-channel FET), but I'm not sure if this > would work. It certainly doesn't work in SPICE, but since when have SPICE > simulations ever accurately modelled real life? :) > > - Microchip AN793 seems to be pretty accurate. The formulae on > <http://www.smps.us/> (buck converter section) seem to agree with the > appnote, and a lot of the other stuff I've looked at says the same thing. > > - I've come up with an inductor value of 68uH for this scenario: > Vin = 6V, Vout = 1.5V, Iout(max) = 1.5A, Iout(min) = 10uA, Fosc=100kHz, > Vripple [permissible ripple] = 10% of 1.5V = 15mV > - Does this seem about right? > > - If I pick a maximum Vout of 3V (i.e. two NiMH cells), I get L=75uH. > If I want to optimise my charger so it charges one or two cells at a > reasonable efficiency (70% or better), should I stick to the lower value, or > pick one in the middle (if one is available)? Or maybe pick the higher value? > Common sense suggests "pick the one in the middle" but I'm not sure how > things are generally done (a lot of the stuff I've found is a bit hazy on > this). > > Thanks. > -- > Phil. | Kitsune: Acorn RiscPC SA202 64M+6G VF+UniPod > .....philpemKILLspam@spam@dsl.pipex.com | Cheetah: Athlon64 3200+ A8VDeluxeV2 1G+180G > http://www.philpem.me.uk/ | Tiger: Toshiba SatPro4600 Celeron700 256M+40G > -

In message <Pine.BSI.4.40.0605261444020.13182-100000KILLspamuniversal.dca.net>> Kevin <.....kbenKILLspam.....dca.net> wrote: > > Vripple [permissible ripple] = 10% of 1.5V = 15mV > > 10% 0f 1.5V is 150mV not 15mV, probably just a typo :) Yep, an extra zero before the percent sign. This: Vripple [permissible ripple] = 10% of 1.5V = 15mV Should read: Vripple [permissible ripple] = 1% of 1.5V = 15mV ^^^^ "Oops"... Thanks for pointing that out, Kevin... Thanks, -- Phil. | Kitsune: Acorn RiscPC SA202 64M+6G VF+UniPod EraseMEphilpemspam_OUTTakeThisOuTdsl.pipex.com | Cheetah: Athlon64 3200+ A8VDeluxeV2 1G+180G http://www.philpem.me.uk/ | Tiger: Toshiba SatPro4600 Celeron700 256M+40G

> > - The Atmel AVR450 appnote is just plain wrong. The section on charge > profiles and charging methods is OK (but check against the Sanyo and > Panasonic charging specs before you go writing any code). I've told other people this.. - Apparently the AVR450 drive circuit doesn't turn the MOSFET on fast {Quote hidden}> enough, so it ends up cooking (there are a few people on the AVRFreaks > forums > saying this). The circuit is a 680R from gate to source, a BC847 (SMD > BC547) > pulling the gate down (E to GND, C to FET gate). The 847 gets its base > drive > from the MCU via a 1k series resistor, and a 10k pulldown on the base side > of > the 1k. Power goes into the source, drain is wired to the inductor/diode > junction. If my description isn't good enough, google for "AVR450 pdf" and > take a look at page 27. You need a better fet driver, I used a driver chip that gave me something like 2A on the gate. Micrel MIC4429BM - I was going to drive the gate of the FET directly from the PIC (and > maybe use an N-channel instead of a P-channel FET), but I'm not sure if > this > would work. It certainly doesn't work in SPICE, but since when have SPICE > simulations ever accurately modelled real life? :) Trust me, it sucks. At least it does when running at 500 kHz. {Quote hidden}> - If I pick a maximum Vout of 3V (i.e. two NiMH cells), I get L=75uH. > If I want to optimise my charger so it charges one or two cells at a > reasonable efficiency (70% or better), should I stick to the lower value, > or > pick one in the middle (if one is available)? Or maybe pick the higher > value? > Common sense suggests "pick the one in the middle" but I'm not sure how > things are generally done (a lot of the stuff I've found is a bit hazy on > this). Most of your losses will end up being diode drops, or resistive I^2R losses. Try to keep those down. If I had it to do again, I would use a conventional switcher chip configured as a constant current source, and then use the micro to switch between current levels as needed, maybe with a Dpot, maybe with a D/A arraingement.