You could start with a dual-primary transformer, with a switch to configure the primaries either in series (for 230VAC nominal in) or parallel (for 115VAC nominal in). Then you can assume that the primary winding's AC voltage might vary from 85VAC to 145VAC. Then select a transformer that will give less than 48V after rectification and smoothing, even for 145VAC in, and design a boost converter.
If you used a transformer with a nominal 24V output, the output would range from about 17.4VAC to about 29.7VAC, which would give roughly 23.2 to 39.2 volts peak after rectification. Subtract half of the ripple voltage's p-p amplitude to get the average VDC.
The ripple voltage's p-p amplitude would be approximately I/2fC, where I = load current, f = AC mains frequency, and C is the value of the smoothing capacitance. With a 2200uF smoothing cap, 60 Hz supply, and 2 A load current, the p-p ripple voltage amplitude would be 7.58 V p-p. The 50 Hz worst case would have 9.1 V p-p ripple amplitude. Half of that would be about 4.55V.
So, with a 24VAC nominal transformer output, the nominal average DC output, after rectification and smoothing, would range from (23.2 - 4.55 =) 18.65 to (39.2 - 4.55 =) 34.65 VDC.
That's probably a bit too low for comfort. And it looks like maybe a 30VAC nominal transformer output might work.
With a 30VAC nominal transformer output: For 85VAC to 145VAC in, output would be about 21.8VAC to about 37.2VAC, giving about 29.4 to 51.2 volts peak after rectification. Subtracting half the ripple amplitude (assuming 2200 uF smoothing cap) gives an average output (i.e. SMPS input) of from 25.61 VDC to 47.4 VDC for 60 Hz mains, or 24.9 VDC to 46.7 VDC for 50 Hz mains.
So, for the SMPS design, you could use Vin = 36V nominal, with Vinmin=25V and Vinmax=48V, and Vout=48V @ 2A. Go to
http://www.linear.com and download the LTspice software. After installing and running it, select File --> Switch Selector Guide. Enter the numbers just given. Select the LT1070HV and click OK. You will see the resulting circuit, and a simulation will automatically run until approx steady state, for the nominal input voltage. It looks like it would be 97.7% efficient. 40 kHz output ripple is less than 15 mV p-p. Scrolling down shows the dissipation for each component. There are only eight external components. Nothing dissipates more than about 0.9 Watt, @ nominal. Nice.
The transformer could be Mouser.com # 546-182N15, a Hammond 182N15 120VA toroidal with dual primaries and secondaries. With 117VAC input, the secondaries, if wired in series, are rated at 30VAC @ 4A. That transformer is 1.68 inches tall and 3.98 inches diameter. You can probably find a similar but cheaper transformer, somewhere.
In place of the 85uH inductor (which has an average current of about 2.7 Amps), you can substitute a Bourns/J.W. Miller 2100HT series 100uH 4.6A toroidal choke (DCR=0.053 Ohms). Those are about 0.86 inch diameter and 0.45 inch thick, and come in both horizontal and vertical mount versions. Mouser part numbers are 542-2100HT-101H-RC (Horiz) and 542-2100HT-101V-RC (Vert). Digikey.com also carries them. Pricing at mouser.com is about $2.44 each qty 1 or $1.69 each qty 10.
If you end up with too much switching noise at the output, you could always add another 1000 uF in parallel with the original 1000 uF output cap, and put one of 2100HT-series toroids between them, to form a lowpass Pi filter.
It looks like the SMPS 120 Hz (or 100 Hz) output ripple might be about 1.7 V p-p when the input is from a 30VAC 60 Hz transfomer output. So you might need to increase the value of the smoothing cap, or try 2x 2200uF.
OK: it looks like adding another 100uH choke and then 1000 uF to gnd, to form a Pi filter at the SMPS output, keeps the output's AC mains ripple amplitude at less than 0.9 V p-p (for 30VAC@60Hz input, at least).
Note that I have not included any transformer regulation effects in any of the above calculations.
- Tom Gootee
**broken link removed**