SMPS for Audio Q's

[weinstro]

Hi,

A couple of basic questions for using SMPS in hifi audio applications:

- What level of noise do you design for? Absolute, or percentage?
- What switch frequency do you design for? I'd like to go high enough to simplify the filtering, but not so high that I start to need exotic snubbers and run into weird EMI issues.
- How do you test for EMI/RFI emissions? Do you care?
- What level of load regulation do you design for?
- What level of current slew rate do you design for? I'm thinking 0.5A/us might be sufficient
- SMPS like to have a minimum current draw; some suggest 20% of rated. What minimum level of current do you design for?

Application, ultimately, would be an active powered loudspeaker based on LM3886 IC's and electronic crossover, roughly similar to Linkwitz Pluto.

Thanks!

[star882]

Building your own mains power supply from scratch is not for a beginner. If you know how to work with high voltage, I would suggest modifying a PC power supply. It's very easy to modify a half bridge design for 24v or 48v output.

[Sch3mat1c]

Originally Posted by weinstro

- What level of noise do you design for? Absolute, or percentage?

Depends on purpose. An audio power amp isn't going to care. Microphone amp maybe. Digital circuitry, not really. High bandwidth analog, yes (radio?).

- What switch frequency do you design for? I'd like to go high enough to simplify the filtering, but not so high that I start to need exotic snubbers and run into weird EMI issues.

20kHz is the minimum to avoid acoustic issues, and 50kHz is a very comfortable frequency without needing large transformers. 100kHz is fairly typical for offline applications; 200kHz and up tends to generate a lot of dV/dt, which means possible common mode noise issues (or switching loss).

Low voltage DC-DC converters can easily run into the MHz; I made a cute LED driver ("joule thief") which steps up a 1.5V AA cell to 3.6V 50mA to run a few LEDs -- it happens to run at 5MHz. More sophisticated controllers and transistors (often using MOSFETs) have excellent efficiency and minimal size in this frequency range.

- How do you test for EMI/RFI emissions? Do you care?

You should care, if even just in principle. It's not easy to test without a spectrum analyzer; your best clues with an oscilloscope will be the "shorted" probe test -- is your ground lead picking up RFI that shouldn't be there? Basically make a loop antenna out of your probe and wave it around, look for anything interesting. I once saw a five-cycle wavelet of 80MHz, 100mVp-p coming off a small converter I built! You're supposed to have "nothing" in the VHF range, where it interferes with radio, television and other more important transmissions (e.g., aviation at 110-130MHz).

- What level of load regulation do you design for?

Whatever I get, which is in the 1% range or better. I recently built a couple-transistor (no controller chip) flyback supply which, thanks to its TL431 error amp, produced an impressively stable 10.2V from 0 to 3A load!

- What level of current slew rate do you design for? I'm thinking 0.5A/us might be sufficient

I've never used dI/dt as a design constraint, only to check against parasitic inductances to see if it's going to fry itself. More amps just means "put the transistors closer together".

- SMPS like to have a minimum current draw; some suggest 20% of rated. What minimum level of current do you design for?

The last two supplies I built, I added no leakage resistors. One, a forward converter using the TL494, turned on, ramped up, then stopped oscillating -- I thought something had gone wrong, until I realized it had stopped because it was unloaded. Adding a load resistor brought it back online. The error amp cut off at the appropriate voltage, so it was in no danger of producing excessive voltage. The other, its control circuit drew about 20mA, more than enough to account for its idle current.

Poorly designed switching supplies may be prone to this, but I have not built one which is.

Tim

[weinstro]

Thanks, Tim

Originally Posted by Sch3mat1c

I've never used dI/dt as a design constraint, only to check against parasitic inductances to see if it's going to fry itself. More amps just means "put the transistors closer together".

Here, I'm more concerned with stability when presented with a 60W transient, like a symbol crash, than with the ability to source a bunch of amps.

[weinstro]

One other question on EMI....

Originally Posted by Sch3mat1c

You should care, if even just in principle. It's not easy to test without a spectrum analyzer; your best clues with an oscilloscope will be the "shorted" probe test -- is your ground lead picking up RFI that shouldn't be there? Basically make a loop antenna out of your probe and wave it around, look for anything interesting. I once saw a five-cycle wavelet of 80MHz, 100mVp-p coming off a small converter I built! You're supposed to have "nothing" in the VHF range, where it interferes with radio, television and other more important transmissions (e.g., aviation at 110-130MHz).

My oscilloscope bandwidth is 10 MHz, not 100. Any suggestions for EMI detection? I'm thinking of interference with a clock radio as my detection method, tuned to either AM or FM bands.

[Sch3mat1c]

Preferably you'd have an electronically tunable radio that covers 1-100MHz or more. Then you sweep the tuning back and forth, and view the amplitude (or AGC signal) on the scope. The tricky part is calibrating it, at which point you might as well buy a real spectrum analyzer.

A clock radio only knows AM and FM, missing the SW band, TV VHF, aircraft, etc. Plus, you'll only hear anything if it's stronger than local stations and modulated (which you could do, by adding a signal to the error amp).

10MHz is really too low for switching supplies, you can't see all the detail, especially if you manage to build something with unusually fast risetimes. How would you ever know, right?

Tim

[weinstro]

Hi Tim,

Originally Posted by Sch3mat1c

A clock radio only knows AM and FM, missing the SW band, TV VHF, aircraft, etc. Plus, you'll only hear anything if it's stronger than local stations and modulated (which you could do, by adding a signal to the error amp).

It's slightly better - a Sony shortwave radio with AM, FM, short and long wave reception. Very sensitive. My amplitude adjustment approach would be to simply move the radio farther/closer to the power supply and listen for issues.

Originally Posted by Sch3mat1c

10MHz is really too low for switching supplies, you can't see all the detail, especially if you manage to build something with unusually fast risetimes. How would you ever know, right?

Agreed, completely, but it's what I have. I'm shooting for "good enough" rather than something that imitates a battery. That's why I started the post with a query for minimum PSU performance specs for hifi. Always good to have a target in mind...

Thanks!

[weinstro]

Hi,

For my project, I'd like to modify an ATX, as this is likely the fastest, most cost effective approach. Output needs are +/- 28 Vdc. I'll also need +/- 15 Vdc, but I can probably derive that from the main supply easily enough. Target load is 3 LM4780 IC's.

I've learned that there are many ATX supplies out there - thousands. But I'm not sure which ones are great to very good vs. junk. I would like one that has:
- A large 120mm or 140mm fan for low noise
- Power > 250W (peak)
- Larger chassis with some room inside
- Low ripple and noise, as this likely means that the supply includes additional filtering components
- Obtainable from eBay for $50, more or less. Used is fine.

Any recommendations? Corsairs look nice....

Should I care if the topology is double forward, half bridge, full bridge, or something else?

Thanks!

[Ar4]

Hello,

I wanted to modify PC power supply too, so I could drive LM4780. I need +/-35V.

So, weinstro, have you built that supply?

[weinstro]

Originally Posted by Ar4

Hello,

I wanted to modify PC power supply too, so I could drive LM4780. I need +/-35V.

So, weinstro, have you built that supply?

Negative. I did figure out that I need ~600 W in order to have the transient response needed for audio. So, I purchased 2 used ATX style supplies, but have not yet reverse engineered them. Maybe two more months.

Regards,

Rob