EEVblog Electronics Community Forum
Electronics => Projects, Designs, and Technical Stuff => Topic started by: sourcecharge on July 02, 2024, 09:53:05 pm
-
Hello,
It's been a while since I posted, but I thought this would be better solved here than anywhere else.
Due to a combination of a function generator repair, and ucurrent upgrades, I found myself wanting to make a shielded function generator isolator that used BNC inputs and outputs.
The last one used a plastic box and the new ucurrent opa189 upgrades was picking up noise every time I turned on the function generator.
I couldn't just relative it out because at about 800khz, the noise was bouncing the ucurrent all over the place.
The catch is that the BNC connectors are grounded to the box, so there is no real reason to used the optocoupler now.
That left using three new high performance opamps that I previously may have over temped when trying to repair a function generator while I was re-flowing them with my hot air solder wand over and over again.
These op amps were THS3095DDR.
There are better op amps that I'm planning to buy, THS3491DDR, but I'm wondering if a irregular behavior of these op amps is normal.
All three THS3095DDR op amps started pulling about 40mA through both positive and negative rails, for about 10 to 15 seconds, and then lowered to the quintessential current of 9.5mA each.
Sometimes it would bounce back between the 9.5mA and 40 mA, but the signal looked fine.
The power down function worked fine, and it would be less than a mA when power was applied and the op amp was disabled, but it would pull the 40mA when enabled, then go back to the 9.5mA.
Admittedly, these chips are on adapter boards to dip pins, put in a bread board upside down with polyimide tape wrapped around the adapter board.
Seemed to work fine when it settled down, and I was able to get a 5 Mhz signal (square and sin) with 5x gain out of them at a full 10Vpp and it could go much higher if I was using sin only and increased the signal input voltage.
They were running at about 60C without a heat sink, with and without a 1k load.
Currently they are all fried due to my own fault, but I was wondering if these op amps normally do this or if those op amps were toast before I fried them by shorting out the pins one way or another.
If its normal, why do they do that?
BTW, on a seperate topic regarding the function generator isolation, if the opamp buffer/gain signal were to be shorted to a 50V source, would the opamp protect the function generator?
-
They look like a great buffer/driver IC. While not having any experience with that particular current feedback opamp, the others I have worked with are prone to oscillation if layout and decoupling are not good and I would suspect that maybe the 40ma is due to some parasitic oscillation which settles out, especially with it being socketed on an adapter board. Regardless the 40ma doesn't sound normal, it shouldn't have a long time constant for settling current unless there is some outside disturbance.
-
The irregular current draw occurs even when the op amp is switched to enabled and input signal was already applied.
After about 10 -15 seconds, it dropped down to quintessential current plus load current for the load resistor which is 1K ohms.
I did forget to mention that I drilled a small slot in the adapter board in the middle of the chip to allow the power pad access to a heat sink, that's why it was upside down.
Also, on the 14pin soic to dip adapter, I am using two unused pins behind the chip, two 20uF surface mounted caps for V+ and V- closest to the supply sides of the chip.
This op amp has a slew rate of about 7kV/us and a bandwidth of about 200Mhz, so maybe you're right, and it's only due to being socketed in the bread board.
Granted the parasitic capacitance is probably high, but if I am able to get a clean 5 MHz square and sinusoidal output at 20Vpp using 5x gain, how much parasitic capacitance could there be?
With that said, could it be settling current if the non-inverting input signal was applied before the op amp was enabled?
-
schematic ?
-
Hi,
Its very simple, nothing fancy.
All caps are within 2cm, all 1K resistors are metal film precision, REF sets the enable and disable voltage reference, and PD is the pin that enables or disables it.
I don't have the rotary switch in the bread board, but it was going to be in the final design shielded box.
(edit: The schematic has an error, the left 39k resistor for the divider on PD should be connected to +Vs not Ground.)
-
You have a rotary switch in the feedback path of a 305MHz/6kV/s current feedback opamp, I don't think that is advisable. Between the socket adapter, and through hole components that is all adding inductance and possible stray capacitance, you are better off sticking with just surface mount as in their 'layout advice' in the data sheet. Also all of the data sheet tests are with a 100R load resistor which will dampen output resonance. These devices really are finicky to such details. Personally I don't really like surface mount, fiddly and hard to handle and see, but in such a case I think one would just have to bite the bullet and go full surface mount and custom pcb.
-
You know, you might be right on the rotary switch...
I changed the switch out to a break before make 2 turn (3 positions) 1 pole, flip switch.
That way, I still have a way to allow 3 different gains.
I changed the design to use the AD8512 op amps in a lead follower configuration using 4 chips, two 16dip soic adapter boards with 2 chips per board.
I had these op amps laying around and didn't use them because of the obvious performance difference between the 3095 and the 8512.
This chip has 2 op amps each, has a full +/-15V supply operation, with only 50mA output, but it's got unity gain.
The first op amp will make a gain outputs of 2, 5, and 10 using one resistor network.
The next 7 op amps will be in unity gain all in parallel, fed by the first op amp, eliminating the need for additional resistors.
Furthermore, the op amp has a Vos of 1mV or less and I going to try to run the in parallel without output balance resistors.
The load should only be a 1k resistor, but the total maximum current output should be around 350mA with only 50mA per op amp.
I will also use the output of the first op amp to feed the input of the TLP352 optocoupler, allowing a somewhat isolated square wave output from rail to rail up to 30Vpp.
The 20V/us of the AD8512 should yield about 2 Mhz, which is less, but faster than the TLP352 anyway.
Maybe later I'll upgrade to the 3491 with the fancy surface mounted custom board.
I think you were right, and after some research, I think the reason why the 3095s were pulling the current was the settling current, but it was probably due to the 2x 30uF CERAMIC surface mounted caps, 2cm away from the source input pins.
I think tantalum capacitors are REQUIRED to stop the settling oscillations, although I might be wrong.
They probably have DFs of about 6% to 10% forming more of a low pass filter next to the Voltage source pins for high frequency noise.
Using the extra pins of the adapter board was good, but I should have use them for the smaller caps.
I only had the large caps on hand though.
And before you ask, here you go.
-
If you are going to parallel opamp buffers, put a resistor in series with each output to prevent the small mismatch in offset voltages leading to large output currents between the buffers because of their low output impedance. It is fairly common for signal generators to have a 50R output impedance, so if all the parallel output resistors added to 50R that would be good.
-
THS3095 on a breadboard?
Read the datasheet and all the literature about current feedback op amps on TI page..
HINT: these amps are VERY sensitive to PCB layout and power decoupling.
-
If you are going to parallel opamp buffers, put a resistor in series with each output to prevent the small mismatch in offset voltages leading to large output currents between the buffers because of their low output impedance. It is fairly common for signal generators to have a 50R output impedance, so if all the parallel output resistors added to 50R that would be good.
Ya, I tried it without them first to see if I could get away without the balance resistors, but it was pulling about +/-100 mA between 7 follower output op amps and 1 lead op amp.
The trick to balance resistors is that you have to balance them very close, any tolerance error between them equals the difference of the amount of current that each paralleled op amp uses.
I had seven 18 ohm resistors that were within 1.3% of each other so I used those.
The initial problem though was noise caused by the lead follower design.
Apparently, the output signal to the other op amps non inverting input was causing so much noise that it was unusable.
Adding one 1k resistor between the output and the non inverting inputs of all the follower op amps took care of that.
Unfortunately, the 8512 has a slew rate only 20V/us.
Obviously, that means a max frequency of 500khz for 20Vpp and 1Mhz for 10Vpp. Sadly, using unity gain, 10Vpp only gave about 400khz.
So, I'm getting back on the horse and not just the bronco that kicked me off three time before, but I'm going for the stallion.
The THS3491, using an adapter board, but soldering it first on a 0.1'' through hole circuit board.
It and the required components are not that expensive, and for what it's meant to protect, I willing to try again.
So I'm going with the original schematic, but with some tweaks.
First, I'm going to keep the non-inverting hold down resistor to 50 ohms, and not 1k.
Second, I need to figure out a way to apply and disconnect the voltage sources with the correct dV/dt for the specified dV/dt within data sheet of the 3095. This way, I can connect the capacitors from PD to V+ if PD is held at 0V.
Third, I going to use mosfets to switch the resistor network of 2, 5 and 10 gains. (see table 9-2 and my schematic)
From ths3491:
Table 9-2. Recommended Resistor Values for Minimum Peaking and Optimal Frequency Response
With RLOAD = 100 Ω
GAIN (V/V)
DDA PACKAGE
RG (Ω) RF (Ω)
2 2.1k 2.1k
5 200 798
10 78.7 704
Fourth, I'm going to use a two BNC outputs with separate 0.9 ohms, or 49.9 ohms in line resistance.
Fifth, I'm going to use 100nF and 100pF surface mounted ceramic caps (0508 & 0603) on the adapter board between unused pins and the source pins making their distance less than 4mm away from the source pins. (see mock up scaled picture)
Sixth, the resistor network will also be surface mounted (0508) on the adapter board between unused pins. (see mock up scaled picture)
Seventh, I'm going to use a tantalum radial 10uF capacitors close to the adapter board, mounted on the through hole circuit board.
And lastly, if going to make sure I put the heat sink on it before I use it. The 3095 was always running hot, and even without a load it was +60C. The THC3491 should probably be higher due to the increase of quintessential current.
Edit:
"Second, I need to figure out a way to apply and disconnect the voltage sources with the correct dV/dt for the specified dV/dt within data sheet of the 3095. This way, I can connect the capacitors from PD to V+ if PD is held at 0V."
I was reading the TI tech bulletin forum for the ths3491, and someone said something abou this specified dV/dt within data sheet of the 3095. I thought there was something special about this, but after second thought, it's probably just refering to the decoupling capacitors. If that's the case, then I'll just use the power supplies to power the op amp down when not in use.
-
So I made a mistake on the design of the mosfet switching.
There is a 30V swing on the gate of a mosfet that has a 15V source.
The mosfet can be changed to a 30v Vgs type, but I'm working on a fix right now.
edit,
I'm just going to use dpdt relays, like the ones inside my function generator.
A total of 3 is all that is needed for a total of $7.53 compared to the price of mosfets now a days.
I keep thinking that they are pennies because I bought so many of them so long ago, I think in terms of mosfets.
-
Just a couple of comments, hardwire the lowest gain setting permanently, you don't want the possibility of complete loss of feedback. For current feedback opamps don't use a feedback resistor between output and '-ve input' higher than that recommended in the datasheet, I think their preferred value is around 1k. Current feedback opamps have dissimilarities with voltage feedback which it is good to be aware of. Just change the 'to ground' resistor value in the feedback path, which means for 3 gains you only need 2 switches.
-
Just a couple of comments, hardwire the lowest gain setting permanently, you don't want the possibility of complete loss of feedback. For current feedback opamps don't use a feedback resistor between output and '-ve input' higher than that recommended in the datasheet, I think their preferred value is around 1k. Current feedback opamps have dissimilarities with voltage feedback which it is good to be aware of. Just change the 'to ground' resistor value in the feedback path, which means for 3 gains you only need 2 switches.
Hardwired and compensated.
The data sheet said 2.1k for both Rf and RG (R1&R2) for gain of 2 for the DDR soic package. The RTG package has it at 976. The graphs show a slight decrease in gain attenuation at higher frequencies with the 976 resistance, but I'm guessing that the lower resistance means more heat generated, hence the difference between the two packages. The RTG package must dissipate heat better. I plan to cut a hole in the adapter board, and solder a flat copper wire stem on the power pad to attach a rather large heat sink to it. I went with 1.2k ohms, and will monitor the temp under load when completed.
I realized that I only needed a gain of 2 and 5 because I'm using a 1 ohm output and a 50 ohm output, so now it only requires one switch, but I'm going with two switches using one for cutoff of both outputs.
Here is the mock up and the schematic.
Edit:
In the picture of the mock up, there are resistors that are soldered between the dip pins. These resistors are on the other side of the board.
2nd EDIT
I changed it back to a 10x gain included.
-
I really like how you have turned the DIP adapter board into an smd breadboard, nice work, I'll need to keep that in mind. :)
-
Thanks, but I sure I'm not the first to do this.
It seems as if the the irregular current draw was due to damage that I caused not by the heat solder wand, but from turning the current amplifiers on with the bench top power supplies.
Apparently there is a maximum of 1 V/us of source turn on voltage slope that if exceeded, it could damage the amplifier.
Adding a current limiting resistor in series before the decoupling capacitor seem to work.
Other than that, I think I ready to start making it.
-
How are you connecting the heatsink pad to a power plane? That is how the device dissipates heat, without it it could cook.
-
I came up with an out of the box idea that I did for the first 3 ths3095 chips.
I measured the pad temp without a load and they were running about 60C.
During startup, they were going up to about 100C.
Now that I'm sure that I know what was the problem with the runaway current at startup, I think the idea still had merit.
I cut a slot in the middle of the adapter board.
This time it will be longer so more copper and route the heat better.
It's about 3.2mm wide by about 14 to 20 mm long, depending on if the board can handle 20mm and how rigid the copper sheet will be when soldered to the pad.
I have 0.5mm thick copper sheet on hand so about 6 layers should do it.
I plan to use two of these heat sinks, sandwiching the copper sheet with 2 screws holding them together.
These are the heatsinks:
https://www.digikey.com/en/products/detail/boyd-laconia-llc/504222B00000G/5833 (https://www.digikey.com/en/products/detail/boyd-laconia-llc/504222B00000G/5833)
DigiKey Part Number HS104-2-ND
Manufacturer Boyd Laconia, LLC
Manufacturer Product Number 504222B00000G
Description HEATSINK TO-220 PWR CLR 1.45"10W (Dual) Aluminum 4.0W @ 40°C Board Level
I already have a lot of these for mosfets with the TO-220 package (irf530 and irf9530)
Here is the mock ups of the front and side view of the copper stem, and the back view with the heat sinks.
I might use more of a T shape for the copper stem to fully support the heatsinks' length of 36.8mm.
This might be more efficient than the thin copper pad with 13 vias on a pcb of which the data sheet recommends.
I'm also thinking about 3D printing stem supports that will protect the pins and make the sheets more rigid in a slot created by the 3d printed supports that will be holding the edges of the board.
On a side note, I'm going to mill off the copper pads and traces on the opposite side of the adapter board that are for tssops packages, leaving only the dip pin pads of the opposite side from the chip. This should remove any stray capacitance from those pad and traces.
-
I love your graphics, what CAD program are you using for them? The pad is normally soldered to its heatsink, the copper plane, so one gets good conductivity from a small area, you would need to use a good heatsink compound, and keep the contact area as flat as possible.
-
Well, thanks for the compliment, it's just paint, I use gimp too.
I wound up frying the chip, literaly. I heated up the copper slug to about 160F and then applied the chip with solder paste under it.
I fryied the chip, and I moved on.
I wound up using the LT1206, the little brother of the LT1210, because it has 60Mhz BW and 1/4 amp output on a to220 package.
I have it all up and running and it's good up to about 5Mhz at -0.1 DB
I got it well cooled, and isolated inside a metal box with mesh in the vents to let air in and out.
It did the trick, and I'm never using those exposed pads again because I rarely make or order pcbs.
Also learned that alot of the newer opamps have a voltage source slew rate maximum. You turn them on too fast, they fry.
Anyways, now a days, I'm making a highside ucurrent spinoff using the 1st stage of an instrumentation amplifier.
I've got some spare 10mohm surface mounted shunts that I thought were bad, but they apparently are fine.
Combined with some low noise dual opamps, I'm able to get a good highside differential voltage measurement from DC to about 200hz with mV being mA.
Any higher though, the output attenuates to unusable measurements.
It's because I used a 100x gain on them.
I've verified that a 10x gain will work with the opa189, up to at least 10khz with no attenuation, and therefore, I'm going to use them (opa2189) with a 100m ohm shunt and 10x gain.
This will allow low series resistance in the current measurement, and by only using the opamps in the 1st stage of an instrumentation amplifier, I don't have to worry too much about balencing the resistors of the subtracting stage.
-
Glad to hear you got something working, those thermal pads under the IC or power device are popular but a pain as they definitely need a custom PCB. :)