Favourite Class D amp for instrumentation purposes

A common mode choke (transformer) at the output of a class D amplifier reduces switching noise enourmously.

You probably know this, but LTC1923 (not cheap) is designed for the task.

If you just need one polarity, something like the LTC3810 maybe.

Yikes. That almost calls for a semi-discrete solution or at least one where the power devices are not on the control chip. Maybe look at the ADN8831? It'll still be pretty small.

If it causes temprature ripple it'll be impossible to get rid of. Must filter that before it gets to the TEC.

SMPS.

Not if it has the power devices on the chip. And only if it's only onesies and twosies so you live off of samples in case they can't deliver some day :-)

Careful. As some point you might hear a "tsssk", a hiss, and then there's a crack in the sensor.

That does a lot of nice stuff, but max Vcc is 5.5. Bummer. More and more, linear parts only work at low voltages.

You could AC couple and DC restore into the pfet gates, which would only take a few tiny parts. That would let you power the h-bridge from a higher voltage but preserve the nice slow gate drive feature.

John

Careful. A slow gate drive when the top FET has to swing to 24V or whatever can quickly result in a hiss, a bang and a puff of smoke.

Thanks, I'm definitely going to be filtering six ways from Sunday. I'll probably suck it up and use cap multipliers, since I've got a lot of power supply voltages, so I can get rid of most of the V_BE drop by adding a bit of extra base bias. That'll cost me almost a watt at maximum output, though, which I'm not crazy about. It's still probably better than class AB. There's no way I know of to get that sort of ripple rejection in as small a board space.

Cheers

Phil Hobbs

That's the leading candidate at the moment. I can filter the thermistor leads pretty vigorously. Any suggestions for FETs?

I have a two-speed control loop--one with about 0.15 Hz BW for the temperature control, which gives the wide tuning range, and a second one with about 500 kHz BW via current-tuning. The current-tuning loop has an extra gain of about 5000 over the temperature loop, because the current tuning sensitivity is so small. The loop locks up using just the temperature, and then the firmware dorks the temperature setpoint until the current loop comes into range. Once that happens, it takes over completely because its loop gain is so much higher than the temperature loop at all frequencies above about 10 microhertz.

The prototype is the one we discussed here late last summer. I built two of them and heterodyned them together on a photodiode. Their relative stability is well within spec above about 0.1 Hz, but they move around way too much at longer times, apparently due to thermal drift.

(I have a brass plaque on my cube wall at this client's place that says, "DC: The Final Frontier".)

Cheers

Phil Hobbs

Yeah, bummer. The datasheet does illustrate a higher voltage application with the addition of a couple of LTC1693 driver chips, but you lose the slow gate drive feature.

I don't know if this is worth considering:

Another engineer I knew designed a circuit that amplified the ripple and fed it back out of phase to smooth the output of his supply. It worked very well. That was at 120 Hz, however.

Just a thought.

John

Thanks, I've done that too. There's also the Kanner Kap, which is sort of the same idea, but using an activated RC approach. You never get as much rejection as with a cap multiplier, though.

Cheers

Phil Hobbs

Careful, sure. Slowing down gate drive can cost a little efficiency but reduce a lot of EMI. The limit is that the switch frequency will have to go down as the edges get slower, so the inductors will have to get bigger.

John

d I

current

has

ptical

fit

, and

tween

ore

s of

e back

revent

t off

the

u'd

rror

is

e it.

y well.

the maxim chip does not have the temperature sensing electronics on the same chip, its just a switching bipolar current source, you have to implement the sensing and control externally

-Lasse

About 20 years ago, we made a control system to support a Princeton Instruments diode array (line scan) CCD camera. The system included a drive for the Peltier cooler. First tried a SGS L296 switching regulator. Even with the best filtering we could muster, still wound up coupling switching noise into the CCD. Wound up going to a linear post regulator and a LOT of attention to grounds.

Condensation - purge the assy with dry Nitrogen. That's what PI did.

Take a look at the eval board:

They use FDC6020 which are N/P channel combos, meaning two FETs per package. Mouser has over 30k of them in stock and they cost well under a buck a pop:

There's also lots of dual FETs from other manufacturers. 6-pin packages save real estate.

Oh that sounds all too familiar except my current tuning loop was in the low tens of keelohoitz. Was a fun project. One where you are a bit sad when it all works nicely because that means it's done and no more work needed on it.

Hang a sign underneath: "In this cubicle you enter sub-DC turf" :-)

SMPS.

Reminds me of a guy who "fixed" his conducted EMI issue. "What did you do?" ... "Put snubber at the gate, gave me 3dB more than needed" ... "At the WHAT?" ... minutes later ... *PHHHHOOMP* :-)

SMPS.

Series gate resistors can be great. You usually need some, just to keep a mosfet from oscillating. Crank that up, and the switching slows down. Nothing wrong with that as long as the dissipation doesn't get silly.

We have one board that uses a mosfet in linear mode. I used a 10K series gate resistor... but The Brat ran a several-inches-long inner-layer trace between the resistor and the gate, so of course it oscillated. Had to add another 100 ohm resistor right at the gate.

John

SMPS.

Ok, a few ohms or maybe a (very wee) bead. But what I really don't understand is when people place the mother of all gate drivers, point-something ohms of Rdson, and then add a 22ohms gate resistor.

Usually I try to get away without, doing nifty layout tricks. I don't even have to explain those much to my layouter anymore. On one PWM coil driver the client's guys ran some mean sequences and then touched the FET. "Why doesn't this get hot?"

What was her punishment? :-)

optical

SMPS.

$10K raise.

John