DAC output help

[juniorhifikit]

Hi all - I'm having occasional problems with the DAC outputs that create the VCA control voltages in a design I've been working on.

What happens is, every now and then upon power-up, a single channel of one of the 4 channel DAC's will start sending out full-rail voltage (5v), or nothing, depending on the data being sent to it. Under normal conditions, the maximum DAC output is 4.096v. The other three channels of the DAC continue functioning fine, which leads me to believe that it's a problem on that particular DAC channel's output, not the input. It seems like the output amp on the DAC chip is failing. The fact that the channel still responds to data seems to indicate that the resistor string of that DAC channel is still functioning, but the amp is shorting internally and sending out the full supply voltage of 5V.

I'm wondering if the wimpy little output amp on the DAC doesn't like being connected to the -3.7V CV bias, and may need to be isolated somehow. In the datasheet it says:

7.6 Application Examples
The MCP4728 device is a rail-to-rail output DAC designed to operate with a VDD range of 2.7V to 5.5V. Its output amplifier of each channel is robust enough to drive common, small-signal loads directly, thus eliminating the cost and size of external buffers for most applications. Since each channel has its own configuration bits for selecting the voltage reference, gain, power-down, etc., the MCP4728 device offers great simplicity and flexibility to use for various DAC applications.

and earlier says:

7.2 Layout Considerations
Inductively-coupled AC transients and digital switching noise from other devices can affect DAC performance and DAC output signal integrity. Careful board layout will minimize these effects. Bench testing has shown that a multi-layer board utilizing a low-inductance ground plane, isolated inputs, isolated outputs and proper decoupling are critical to achieving good DAC performance.
Separate digital and analog ground planes are recommended. In this case, the VSS pin and the ground pins of the VDD capacitors of the MCP4728 should be terminated to the analog ground plane.

Or possibly if the LM7912 regulator that's providing the bias voltage (who's supply voltage is derived from the -18V leg of the PSU) is sending a spike upon power-up.

Here's an overview schematic of my circuit (sorry for the cut-off image - try reloading in a new window). Any thoughts?

[mediatechnology]

It does appear to be some kind of latch-up.

You might want to consider how the supplies sequence. Odd that it only occurs on one output. Is it the same output after substituting the MPC4728?

[juniorhifikit]

No, it's a random problem. I have 4 DAC's for 16 channels, and it's happened twice now - once on chnl 9 and once on chnl 16. I can't get the offending channel to "un-latch", and have to replace the DAC. Perhaps the bias regulator should be something like an LM2991 with an on-off pin. Is there any simple method of delaying the power-up of the existing LM7912, or give it a "soft start"?

[mediatechnology]

I see that they recommend a 5K minimum load.

I also have to wonder if the outputs are becoming destructively reverse-biased on power-up. There are analog switches between the OA outputs and pins.

As a start, you might want to consider placing diodes at the output pins to clamp them at ~ -600mV and scaling the output resistor values.

Take one of the parts you've removed and check the resistance between the blown output and Vss. I think you'll find it to be Ohmic.

[JR.]

I recognize that circuit over by the VCA But the inputs are reversed on the first opamp.

The pins on the DAC are all rated for +/- 0.3 volt of their rails and if -12v supply comes up before 5V or falls after, the dac output could have a few mA being sucked out toward -12v clearly more than -.3v

I see a few ways to address this. Diode clamps, which may be inconvenient for the voltage range and parts intensive.

Sequence the power supplies,, i.e -12 is tracking the +5V so always -2.5x the 5V rail.

Third, rethink the VCA buffer. While running the DAC and resistor to -12V through a virtual earth inverting buffer would pretty much insure that the DAC never sees -12v, it means adding another whole opamp per VCA;

Option 4 instead of using one 6.34k resistor from the junction of the dac outpout and -12v bias string, connect the DAC and bias string to C8 through two separate 12.6 k resistors. This way the dac is not getting creamed by the -12V supply, Cost is only a few extra resistors and maybe extra caps to filter the -12V rail (btw making that a clean tracking rail wouldn't hurt either if this bias is all it does.)

I vote for door number 4 because it should protect your silicon and still use that clever VCA control circuit.

JR

[mediatechnology]

+1 John.

With analog switches in series with the op-amp output, the OA is likely not there to help sink the current if, during intialization, they're open.
That makes the current path likely flow through the ESD diode and/or some parasitic to the substrate.
It's also a 1 Ohm switch (IIRC) so it must have a lot of area and, when open, capacitance from the output to gate. That could be a path too.
I've seen ESD diodes fail with low-voltage high-current events yet survive high-voltage low-current events.
In this circuit the peak currents can't be very high but it's obvious something is destructive.
Measuring the resistance with a low-current DVM from defective output (and a good one) to Vss and Vdd may tell us something.
At some point you get enough metal blown around that you can't tell much from a pin's I-V curve.

[juniorhifikit]

Option 4 instead of using one 6.34k resistor from the junction of the dac outpout and -12v bias string, connect the DAC and bias string to C8 through two separate 12.6 k resistors. This way the dac is not getting creamed by the -12V supply, Cost is only a few extra resistors and maybe extra caps to filter the -12V rail (btw making that a clean tracking rail wouldn't hurt either if this bias is all it does.)

I vote for door number 4 because it should protect your silicon and still use that clever VCA control circuit.

JR

Oh I DEFINITELY want to keep using that clever bipolar CV circuit . Currently, the -12V bias is super clean, and the VCA control is accurate to about 0.01dB

So you're saying have two 12K6 resistors in parallel, like this? My only concern is that this might change the VCA's scale, which is derived from the entire series resistance, based on the 4.1V DAC output.

[juniorhifikit]

+1 John.

With analog switches in series with the op-amp output, the OA is likely not there to help sink the current if, during intialization, they're open.
That makes the current path likely flow through the ESD diode and/or some parasitic to the substrate.
It's also a 1 Ohm switch (IIRC) so it must have a lot of area and, when open, capacitance from the output to gate. That could be a path too.
I've seen ESD diodes fail with low-voltage high-current events yet survive high-voltage low-current events.
In this circuit the peak currents can't be very high but it's obvious something is destructive.
Measuring the resistance with a low-current DVM from defective output (and a good one) to Vss and Vdd may tell us something.
At some point you get enough metal blown around that you can't tell much from a pin's I-V curve.

I'll measure that tomorrow if I still have any of the fried chips around. They're rather hard to handle, as they're MSOP10 package. Can't drink any coffee before soldering

[JR.]

The DAC can drive 15ma into a short when it's 5V power is up, but if -12v comes up first, or 5V collapses first, the DAC has a few ma being sucked out of it, with no active power so stray internal paths and possible failure can ensue.

Look at my suggestion #4.

JR

[juniorhifikit]

The DAC can drive 15ma into a short when it's 5V power is up, but if -12v comes up first, or 5V collapses first, the DAC has a few ma being sucked out of it, with no active power so stray internal paths and possible failure can ensue.

Look at my suggestion #4.

JR

like I drew - with two 12K6 resistors in parallel instead of one 6K34?