Signal Generator Max Current problem

[CervejaMan]

Hey there,

First post here on the forum, pretty exciting !

So, in my company we have an internal project which started out as a portable +-10V power supply with precise 100mV steps. This project has changed alot in specs, ending up becoming a 0-5V power supply with 100mV steps (more like 0.8 - 5V).

After the design was done, higher ups started asking if it was possible for the power supply to output square waves. After implementing simple "turn-on" and "turn-off" logic with mosfets, they decided that having triangular waves was a must as well. This last requirement lead to a complete redesign of the project, switching from power supply to signal generator.

When it was a power supply, the circuit was an adjustable output voltage regulator with the two feedback resistors + a third resistor connected to the output of a DAC. Changing the output of the DAC changed the output of the voltage regulator and the voltage steps were achieved.

Since the minimum output value of the voltage regulator was 0.8V, it wasn't able to output a triangular wave. The found solution was to just use the DAC as the "voltage source", buffered with a precision opamp. This is where the design gets murky, because alot of money and time was already spent getting to the previous solution, I tried to reuse as many parts as I could.

The DAC (MCP4922) only outputs positive values, so the two channels control the two outputs. For example, if the required voltage level is positive, then that voltage level is applied to channel A while keeping channel B at 0 and when a negative voltage level is required the voltage level is applied to channel B while keeping channel A at 0. This posed a problem for the opamp, since it couldn't really output 0V when its VS was 0V. The minimum value was 170mV. This was completly ruining the 100mV step requirement, so a voltage inverter to generate -5V was soldered to the PCB, cutting the ground connection to VS and replacing it with -5V. This worked out, even when the output of the voltage inverter rises to -4.6V at turn on.

Basically, the design works and the higher ups are happy. It's a sandwich with an Arduino Uno at the bottom, a custom board in the middle and an RGB LCD Shield from Adafruit on top to make the user interface. But, I'm totally not satisfied with the result, it just feels and looks bad from any angle you see it. The truth is that I don't have more knowledge to make it right. I'm using BNC connectors as the output, which isn't ideal for "power supply" applications, but since we have a lab and have many of those cables laying around it seemed good. And, by using a following opamp, the output current is somewhat low. We have some loads here that need high peaks of current (100-250mA) at startup but drop considerably after being turned on (<1mA). I want the design to be able to withstand those high peaks without droping too much in the voltage level (taking into account the drops on the cables, etc) and also wanted it to be able to leave the output in HiZ.

I've seen for other opamps that having multiple units in parallel would make them able to supply more current, but they are expensive and Mouser doesn't have the version with 4 amps available. Higher output capacitance to be able to withstand the peak current could be an option, but they would just ruin the dynamic performance. Would something like a series pass transistor work ?

For the HiZ requirement, I've been looking at how the ADALM1000 from Analog Devices and they have these electrical switches (example ADG719BRTZ). Could the solution be something like this? That one in particular does not work, it can't handle the currents that we're trying to achieve, but would the solution be around something like that?

PS: I graduated college 2 years ago and started working one month after defending my thesis. I've been working alone in the electronics department, so everything I know has been self-taught or taught at college. I'm attaching the schematics and if someone could give me some tips/feedback on them I would be really appreciated! This version implements the atmega processor on the board to try and not use an arduino board since they're expensive.

[tatus1969]

Which bandwidth do you need? From your thread I see +-0.8 ~ +-5V and 0 ~ +-250mA peak, right?

The ADA4807 will typically deliver ~4V at an output current of 70mA (datasheet fig.53), so that would already violate your specs.

The -5V supply will also not be able to source high current, maybe 50mA. That is another limit, but not for transients of course. These can be supplied from the capacitors.

You say your load has transient current peaks. Is that because some capacitor needs to be charged? That could be a problem, since opamps do not like driving capacitive loads, this harms their stability.

Using the outer contact of the BNC for your second output: of course possible when you make sure nothing is accidentally shorted to ground, but because the outer contact is originally meant to be ground / shield, maybe a pitfall in the future. Better use 2 BNCs, one for each output? This would allow your outputs to be shielded, using two coax cables.

[CervejaMan]

The new bandwidth is +-4.5V. This is the last requirement, since it's powered by USB or 3xAA batteries. If it could be stretched to +-5V that would be ideal.

The chosen -5V supply is bad, but it's only there for the opamp to be able to output the 0V (1.52mV more like it). That opamp should be sinking and not sourcing current, since it's the lowest voltage from the pair. So I don't understand why it is falling in voltage level.

And yes, our loads act like capacitors: until they're charged they draw a massive ammount of current, but then they drop to very low values a couple of hundred miliseconds after. The idea is to drop the opamps, or get another way to drive the output that would be better (current wise) and less expensive. Attached is a screenshot of a current profile of one of our loads.

And yes, the BNC connectors aren't ideal and using the shielding for the second signal isn't that good but it beats having four banana connectors on the board because it would take too much space and it would be confusing for the target audiance (sales reps) ... If there was another solution I'd be happy to hear it!

[tatus1969]

The new bandwidth is +-4.5V. This is the last requirement, since it's powered by USB or 3xAA batteries. If it could be stretched to +-5V that would be ideal.

The term bandwidth relates to frequencies. You were talking about triangular waves as well, and the opamp that you chose is quite fast (180MHz GBW). In other words, how fast will your triangle be? How fast should the transition be for the rectangular shape? And for your load, how much capacitance does it have?

Again, your chosen opamps can only deliver +-4V at 70mA. Running on batteries that would be ~3.5V. After the transient, when current has dropped to that 1mA, they provide the full voltage. But again, opamp and large capacitive load is a nogo.

The chosen -5V supply is bad, but it's only there for the opamp to be able to output the 0V (1.52mV more like it). That opamp should be sinking and not sourcing current, since it's the lowest voltage from the pair. So I don't understand why it is falling in voltage level.

When one opamp generates 4V and the other generates 1V for example, the load current comes from the +5V, goes through the first opamp, then the load, then the second opamp, then into the -5V supply. So all four need to be built to your desired current rating. Again, as you seem to need it only for transients, no worries but make sure both supply's output caps can deliver it long enough.

And yes, our loads act like capacitors: until they're charged they draw a massive ammount of current, but then they drop to very low values a couple of hundred miliseconds after. The idea is to drop the opamps, or get another way to drive the output that would be better (current wise) and less expensive. Attached is a screenshot of a current profile of one of our loads.

Depending on the required bandwidth, you could go with a small audio amplifier. They can handle larger currents and, if you add 1 to 5 Ohms in series to the output, will be rock stable.

And yes, the BNC connectors aren't ideal and using the shielding for the second signal isn't that good but it beats having four banana connectors on the board because it would take too much space and it would be confusing for the target audiance (sales reps) ... If there was another solution I'd be happy to hear it!

Work single-ended. Use only one opamp to create +-5V, and use the ground as ground.

[CervejaMan]

The frequency of the wave signals we use is usually under 1Hz. Just last week we used a trapezoidal wave that lasted for 2.5hours (one single period of the wave). So bandwith is not a problem at all.

I chose that opamp because it was rail-to-rail and according to the simulation it didn't drop below 4.5V when the 25mA were at the output. Other opamps that I chose dropped to 4.3 or less, with the 5V positive supply.

The main requirement is precision output voltage: with this design we have a +-2mV error on each voltage step, open circuit voltage. But the maximum error under the maximum output current (25mA) is 15mV. This is measured across the load, since across the conector the voltage level only drops 1 or 2 mV. The majority of the voltage drop is on the cables.

I like the idea of working single ended but since the board is being powered through USB or 3xAA batteries, how could I generate a negative voltage level with high enough current capability to power the opamp? I know some DC-DC converters, even at +-15V but they only supply like 20mA tops. I'm clearly missing something.

[tatus1969]

I chose that opamp because it was rail-to-rail and according to the simulation it didn't drop below 4.5V when the 25mA were at the output. Other opamps that I chose dropped to 4.3 or less, with the 5V positive supply.

So your requirement is 25mA and not 250mA? That is doable with that opamp.

The main requirement is precision output voltage: with this design we have a +-2mV error on each voltage step, open circuit voltage. But the maximum error under the maximum output current (25mA) is 15mV. This is measured across the load, since across the conector the voltage level only drops 1 or 2 mV. The majority of the voltage drop is on the cables.

Have you thought about additional sensing wires directly at the load?

I like the idea of working single ended but since the board is being powered through USB or 3xAA batteries, how could I generate a negative voltage level with high enough current capability to power the opamp? I know some DC-DC converters, even at +-15V but they only supply like 20mA tops. I'm clearly missing something.

If the -5V that you have is not sufficient (it should deliver some -4.6V at -25mA when supplied from 5V), then you could consider using an inductor-based voltage inverter instead. For example LT1611, that is advertised being low noise as well.

[DaJMasta]

I don't have experience with this, but what about sticking an audio power amplifier on the end of your current opamp output.  Either with discrete transistors or an integrated module - it should give you a lot of current handling capacity and low end bandwidth - though certain designs may not go as low as you're thinking.

It's likely not going to be as precise as your opamp's output, but you may not need it, and using a dedicated IC or a commercial amp could give you a good baseline accuracy - maybe then tweaking the output for calibration.

[Kleinstein]

There are a few OPs that can drive something like 200 mA close to the rails. When you need a kind of Rail-Rail operation, the choice is not that large with audio amplifiers. Except for maybe a slightly lower price there is not much advantage over an OP. There are even OPs that include an output enable - so one might get away without extra high current switches.

To really get a 0-5 V output, one would need more than just the USB or 3 AA Cells for supply. Worst case this only something like 3.6-4.5 V. One might get away with a - 1 V and maybe an extra 1 V over the existing supply. The isolation from capacitive load may require something like an extra 500 mV drop (e.g. resistor + inductor), but there is no big difference in making an extra 1 V or 0.5 V.

As USB is often noisy and may be a cause of ground loops, one might consider using a custom made isolated DC/DC anyway, so one could start at something like -1 V and +6 V or similar for the output stage.

[CervejaMan]

In my previous reply I've said 25mA as the max current output because that was the current value that wouldn't compromise much the output. The main requirement still is an accurate voltage level like 1.1V and 1.2V, etc. I just wanted to increase the output current capability of the circuit, knowing that the accuracy of the output voltage will be compromised due to voltage drops on cables and connectors.

Going back to what tatus said, adjusting the output caps of the voltage regulator is enough? No right ? Because the opamp is supplying the current and it can't source, for example, 100mA of current in a pulse correct? I know that the voltage accuracy during the pulse will drop, but if the circuit can keep it as close as possible it would be great. This would allow for bigger loads to be driven using this circuit.

You also said that opamps don't like capacitive loads, that it would be better to have an audio amplifier, but adding 1 to 5 ohms to the path of the signal will decrease the accuracy of the output voltage and it may deviate alot from the +-2mV accuracy we have now. Is this the only option?

[tatus1969]

You also said that opamps don't like capacitive loads, that it would be better to have an audio amplifier, but adding 1 to 5 ohms to the path of the signal will decrease the accuracy of the output voltage and it may deviate alot from the +-2mV accuracy we have now. Is this the only option?

Stepping back on this one after I read the datasheet of your chosen opamp. They specifically state in the datasheet that it is stable with unlimited capacitance at its output (page 26). This is for G=2, but you are using G=1 which theoretically could make stability worse but I don't think there will be a problem. Just keep opamps and capacitance at their output in your mind for next designs to come.

[David Hess]

There are bench power supplies which can do what you want.  They include a small current sink stage in parallel with the output allowing the voltage across the output capacitor to be pulled down quickly.

You can use a bunch of operational amplifiers in parallel but it takes a little bit of finesse and if using a dual or quad, you need to watch out for the maximum power dissipation for the whole package.  See the schematic below for an example.

Read Linear Technology application note 18,  Power Gain Stages for Monolithic Amplifiers, for some alternative ideas and the need for frequency compensation.

Consider using a diamond buffer inside of the operational amplifier's feedback loop.  The old National LH0002 is a good example.

[dom0]

At these very low bandwidths you could just add a common emitter output stage (not a common collector circuit like a complementary emitter follower or diamond buffer) to any R2R (precision?) op. This relaxes several requirements, notably high current and R2R capability in one op. At these currents (25 mA at 4.5 V with 5 V supply?) something like a BC337 would already work.

[David Hess]

At these very low bandwidths you could just add a common emitter output stage (not a common collector circuit like a complementary emitter follower or diamond buffer) to any R2R (precision?) op. This relaxes several requirements, notably high current and R2R capability in one op. At these currents (25 mA at 4.5 V with 5 V supply?) something like a BC337 would already work.

The application note I linked gives a couple examples of common emitter output stages.  It is not their speed which is a problem but their voltage gain which needs to be controlled and the extra complexity of biasing.

Figure 2 shows probably the simplest configuration; replace the LT1010 with the operational amplifier of your choice.  The operational amplifier output is grounded and the current through the supply pins is used to bias and control the common emitter output stage.  Current mode feedback is used back to the output of the operational amplifier through the resistor divider to control the gain of the output stage.  Obviously this will only work with single parts and not duals or quads.

Using an integrated operational amplifier like this with 2 high impedance inputs, one low impedance input, and a differential output is an old technique for getting better performance out of an operational amplifier.  The oldest example I know of where this was used is the Tektronix 7A29.  Add a pair if cascode transistors and it allows a low voltage operational amplifier to control high voltages.