Power = Supply=20 Design

A power supply design blog by Jerrold Foutz.

| SMPS = Technology=20 Home | Blog = Home |=20 About Blog = |

=C2=AB=20 TI Power Supply Design Seminar 2004-05 | Main | How To=20 Layout a Power Supply =C2=BB

January 04, 2005

How To Design a Power Supply

"And if the power supply is not designed properly, it may well = attract too=20 much attention by not working." -- Ed Walker

One of the problems most new circuit designers encounter when they = design=20 their first power supply is how to go about it. As it turns out, there = are a=20 near infinite ways to do it. Left to their own devices a new designer = will=20 usually do a poor job of it. Those in luck, will have an experienced and = vigilant mentor to help them through the process. But for those that = don't,=20 papers like Topic 3, Design Review: A Step-By-Step Approach to AC=20 Line-Powered Converters, in the "Texas=20 Instuments 2004/05 Power Supply Design Seminar are invaluable.

My favorite quote from the author, Ed Walker, is the understatement = taken=20 from his introduction:

"And if the power supply is not designed properly, it may = well=20 attract too much attention by not working."

A power supply is primarily a buffer or interface between a power = source and=20 a load, which, without the power supply would be incompatible. It also = has to=20 interface with an environment. I like that the paper recognized this by=20 discussing load, power source, safety, EMI, mechanical, and thermal = design right=20 at the beginning in the requirements section of the overview.

Then comes defining the power train consisting of: Topology, Input=20 Configuration, Transformer Design, Voltage-Mode or Current-Mode Control, = and=20 Output Filter Design.

The paper throws the control mode (voltage-mode or current-mode = control) in=20 the middle of this section, which implies you should mix control in with = the=20 power train. My own approach gets the power train working first, and = then goes=20 on to design the control. However, no harm done, and in truth, design is = an=20 iterative process, not a linear flow. Since you need to sense current = for=20 current-mode control and this may affect the power train details, = perhaps it=20 does belong here.

What the paper does not do, thank goodness, is to throw everything = into a=20 SPICE circuit schematic and hope for the best. I have seen enough new = designers=20 do this that I think it is the approach taught in school. Disaster = usually=20 awaits those that take the SPICE total-schematic approach for initial = power=20 supply design.

Next the paper describes the detailed design, called the plan of = record in=20 the paper. Each topic is filled with the expected design equations, part = trade-offs including cost, tables, and figures.

The sequence is logical:

Define Requirements
Define Topology and Control Method
Define Input Section
Transformer Turns Ratio
Output Inductor
Output Capacitors
Output Rectifier Snubbers
Output Rectifiers
Power Switch FETs
IC PWM Controller
Setting Up the Loop
A Few Optional Ideas

The paper ends with the test results and the references.

Ok, well and good. But how do experienced power supply designers = really=20 design? Some continue to do it as in the paper, but the process gets = modified=20 over time as it is influenced by company policies, the designer's = preferences,=20 and other constraints.

I define a good designer as one whose designs go into the field and = meet=20 their performance requirements and don't fail, even in abnormal use. = After=20 performance requirements are met, field reliability rules. No two = designers have=20 the same approach.

One designer I knew who met this criteria started with a drawer = filled with=20 schematics that had been proved reliable in the field. When assigned a = circuit,=20 he found the closest schematic, had a breadboard built, modified the = breadboard=20 until it met the performance requirements, and then did the most = complete and=20 prolonged testing of it I ever witnessed. (Those that start with vendor=20 reference designs follow a similar path, although the design usually has = not=20 been proved reliable in the field.) His designs did not fail in the = field.

How did he come by this approach? He came from a company who first = gave the=20 design to one engineer and when he was finished, gave the laboratory = notebook,=20 schematic and breadboard to a second engineer to perform a design = review. If the=20 circuit proved itself in the field, both got credit, if it failed in the = field,=20 the reviewer could expect a period of poor performance reports and no = raises.=20 Strong motivation to make sure your fellow engineer's circuit worked. = His=20 approach was logical in his initial design environment and worked very = well in=20 other design environments.

My own approach was formed by a design environment where my power = supplies=20 had to go into a computer on circuit boards the same form factor as the = logic=20 boards, which changed form factor and slot pitch for each computer. The = only=20 variable was how many logic slots the power supply occupied. You could = not get a=20 power supply in a single logic board slot, but keeping the slots used to = the=20 minimum was critical. The capacitor tubes controlled the minimum slot=20 consumption and their diameter in turn determined the maximum height the = magnetics could take without increasing the number of slots taken by the = power=20 supply.

I started each design with a tool kit of balsa wood, wooden dowels, = and other=20 modeling material and tools that let me mechanically do the layout of = the number=20 of capacitor tubes and magnetics. The envelope of the maximum magnetics = height=20 controlled the cores that could be used. The power train was laid out on = the=20 board like a jig saw puzzle. Then a design was created that fit the = layout.=20 Later I replaced the tool kit with a desk top PC CAD program. The input = EMI=20 filter was designed and lain out first (since doing it last often = resulted in an=20 EMI filter bigger than the rest of the power supply). Then the Middlebrook = Criteria=20 was used to fit and design the output filter. The rest of the design was = then=20 juggled to make every thing else work. This unusual design method = worked,=20 producing designs that beat the best watts/per/cubic-inch = state-of-the-art at=20 the time.

Other designers start with equations and solve the equations with = circuit=20 elements.

And many use an approach similar to that in the paper.

The point is that you need a starting point such as a reference = design, a=20 previous design you have done, or a design procedure as given in the = paper. With=20 experience, you will develop your own methods. In my experience, no two=20 experienced designers reach the goal of performance with field = reliability in=20 the same way. What counts is not how you reach it but that you reach it = in a way=20 that works efficiently for you. But you have to start some place, and = papers=20 like this serve the need.

Here is the detailed bibliography information and abstract. After the = 2004/05=20 seminar is complete it will probably appear on the Texas Instrument website. Until then you = might be=20 able to request a copy from the author or a TI field engineer.

Reference: Walker, Ed, Design Review: A = Step-By-Step=20 Approach to AC Line-Powered Converters, Texas Instruments 2004/05 = Power=20 Supply Design Seminar, SEM1600, pp. 3-1 to 3-25. 25 pages, 19 figures, = 10=20 tables, 7 references, 1 appendix.

Author Abstract: "An offline, three-output, 150-W = forward=20 converter is used as an example to illustrate the design process for = typical=20 isolated converters. This example emphasizes the basics with a = double-ended=20 forward topology using coupled inductors for output accuracy. Design = issues and=20 trade-off decisions to optimize power efficiency while keeping costs to = a=20 minimum are highlighted. Finally, the presentation of measured = performance=20 results confirms the design process."

Posted by Jerrold Foutz at January 4, 2005 02:58=20 PM