Debugging differential charge amplifier

[SethGI]

Hi everyone,

I realize this is a long post, but I wanted to make sure I'm giving all the details possible so people can help me.

I'm working on a DAQ for a piezoelectric hydrophone (Teledyne TC4013, 3nF ESC element). Because it's a piezoelectric sensor, a charge amplifier makes the most since. Because I'm using it to drive an ADC (Analog Devices ADAR7251) it needs to be fully differential. Further, I only have a single 3V3 supply. So, in short, the amplifier needs to be a charge amplifier which does single-ended to fully differential conversion, on a single supply.

I passed this bit of the design off to a (very competent) peer, but now I've assembled the board and am struggling to get it to work. I just can't produce an output signal with the hydrophone as an input. I list my debugging steps below. I'm hoping for some debugging tips as we're both stumped by this.

The amplifier is designed in two stages: a charge pre-amplifier, and then a normal voltage opamp to do a bunch more amplification. Both stages use a TI THS4551 opamp. I've attached schematics from both the preamp and 2nd stage, although I'm quite confident the problem is in the preamp (then again, if I was never wrong I wouldn't be posting here in the first place).


Here are all the measurements I could take:

-Vdc=125mV at the positive hydrophone terminal (measured at R_2). Here, using the hydrophone and pinger I can measure a signal which is ~20mVpp at 20KHz.

I couldn't detect that signal anywhere else in the system, so all I can report further downstream is the DC voltages at various locations.

-V_cm = 1.9V
For the preamp:
-V_out+ ~= 10mV
-V_out-   ~= 3.3V

For the 2nd stage:
-V_out+  ~= 3.2V
-V_out- ~= 20mV

Not being an expert, those voltages seem strange. With no detectable signal passing through, shouldn't all the output voltages be at Vcm? Then again, maybe that little voltage at the input to the amplifier is just getting blown up and saturating the opamp (is this the problem?).

So far, these have been my debugging steps:

-Using a (voltage) function generator to pass a known signal through the system. It generally works, with some small clipping errors that are (I believe) irrelevant to the current problem I'm having. If that persists and I can't fix it, it'll be it's on post.
-I removed the capacitor from the feedback loop in the preamplifier, so it's just a bog-standard opamp circuit. Still no output detectable.
-I decreased the capacitance of the capacitor in the feedback loop of the preamp to 18pF (the smallest which I had on hand), and still no output detectable.

Please let me know if there's anything else I can provide! I've really been banging my head against a wall on this one and am looking for any guidance at all as to what could be going wrong.


Thanks in advance!

-Seth

[duak]

Seth,

to confirm, you have 125 mV DC on the input connector J1?  Isn't the differential DC gain about 3000X?  It looks like the 1st stage amp is driven hard into saturation.  What are the output voltages when J1 is shorted out?  Perhaps temporarily reduce R1 & R4 to 10K to get a handle on things.

[magic]

Just a side remark, why bother with all that differential stuff if you could build a single-ended amp and convert to differential at the very end?

I'm not familiar with the stuff you use but piezo sounds like something very high impedance and I presume your diffamp chip has bipolar inputs. Where is the bias current supposed to go?

[SethGI]

 

Seth,

to confirm, you have 125 mV DC on the input connector J1?  Isn't the differential DC gain about 3000X?  It looks like the 1st stage amp is driven hard into saturation.  What are the output voltages when J1 is shorted out?  Perhaps temporarily reduce R1 & R4 to 10K to get a handle on things.

Thanks! I'll try that to when I can get back to the lab tomorrow. Eventually, that degree of amplification will be necessary, but good call on reducing it for now.

[duak]

When I skimmed over the data sheet last night I missed seeing the input bias current: 0.55 to 1.5 uA.  This will cause a significant voltage offset across the input because the '+' input current goes right to ground whereas the '-' input current must go through R1, a 1M0 resistor.  I think you'll have to reconsider the choice of devices and topology and use a MOS input device here.

[dmendesf]

Probably best way would be to use a signal transformer (like a 10k:300R) to couple the signal to the first stage. Center tape of secondary goes to ground. This will make the circuit fully differential, give a path to ground for the offset current and lower the impedance of the source.

[IDEngineer]

Thanks! I'll try that to when I can get back to the lab tomorrow. Eventually, that degree of amplification will be necessary, but good call on reducing it for now.

x2 on that suggestion. VERY early in my career, an Applications Engineer from National, visiting our site, taught me to drop gain on op amps "until you prove it's actually working as an amplifier". That little nugget of wisdom has saved me countless hours.

[SethGI]

Just a side remark, why bother with all that differential stuff if you could build a single-ended amp and convert to differential at the very end?

I'm not familiar with the stuff you use but piezo sounds like something very high impedance and I presume your diffamp chip has bipolar inputs. Where is the bias current supposed to go?

That's fair... again I didn't design this analog front-end (I'm more or less the digital guy) so I'm not exactly sure why the first stage is differential. I guess we could replace that chip with a single-ended one, but the board has already been made so we'd have to find one with a suitable pinout.

I see your point about the bias current not having anywhere to go. I tried adding a 1M resistor in parallel with the hydrophone to give it a discharge path, but that didn't seem to change anything.

[SethGI]

When I skimmed over the data sheet last night I missed seeing the input bias current: 0.55 to 1.5 uA.  This will cause a significant voltage offset across the input because the '+' input current goes right to ground whereas the '-' input current must go through R1, a 1M0 resistor.  I think you'll have to reconsider the choice of devices and topology and use a MOS input device here.

Oh, very fair point. Huh, I'll look into that. So if I were to substitute for an opamp with much lower bias current (i.e. TI TLV4120 w./ 0.3pA input bias) could I get away with minimal other changes?

Thanks for the help!

[SethGI]

Probably best way would be to use a signal transformer (like a 10k:300R) to couple the signal to the first stage. Center tape of secondary goes to ground. This will make the circuit fully differential, give a path to ground for the offset current and lower the impedance of the source.

Good suggestion. I'll look into that as well.

[magic]

Huh, I'll look into that. So if I were to substitute for an opamp with much lower bias current (i.e. TI TLV4120 w./ 0.3pA input bias) could I get away with minimal other changes?

That sounds like a CMOS device so have a look if noise specs are satisfactory, particularly at lower frequencies if you care about them. If you need less noise, JFET opamps could deliver that.

[duak]

Here's a link to a design from Analog Devices: https://www.analog.com/media/en/reference-design-documentation/reference-designs/CN0350.pdf

I'm not privy to the transducer's specs or the overall performance targets so it's difficult to give detailed design advice.  I don't know if there are other devices that can utilize the same footprint.  It should be possible to kludge in an alternate device to test out changes so the PCB doesn't have to be respun right away.

Without knowing what the application is, I'll bet that any circuit that operates on only 3.3 V without some transient input overload protection could be overloaded.  I don't think a piezo can generate enough current to damage the input device, but a sufficiently large transient could cause the circuit to block up and take some time for recovery.

[SethGI]

Thank you all so much for the help... this forum never ceases to amaze me.

Here's a link to a design from Analog Devices: https://www.analog.com/media/en/reference-design-documentation/reference-designs/CN0350.pdf

I'm not privy to the transducer's specs or the overall performance targets so it's difficult to give detailed design advice.  I don't know if there are other devices that can utilize the same footprint.  It should be possible to kludge in an alternate device to test out changes so the PCB doesn't have to be respun right away.

Without knowing what the application is, I'll bet that any circuit that operates on only 3.3 V without some transient input overload protection could be overloaded.  I don't think a piezo can generate enough current to damage the input device, but a sufficiently large transient could cause the circuit to block up and take some time for recovery.

I also didn't think the piezo could reasonably damage it, and we're taking lots of steps to isolate this board (it's in it's own enclosure on it's own power supply). So, I'll look into transient protection on a future revision of the board, but for now I'll leave it as is.

Huh, I'll look into that. So if I were to substitute for an opamp with much lower bias current (i.e. TI TLV4120 w./ 0.3pA input bias) could I get away with minimal other changes?

That sounds like a CMOS device so have a look if noise specs are satisfactory, particularly at lower frequencies if you care about them. If you need less noise, JFET opamps could deliver that.

Does something like http://www.ti.com/lit/ds/symlink/ths4121.pdf seem like a good choice. It is a drop-in relacement (same footprint) so if I could get away with that, it'd be ideal. I have to respin the board at some point to fix a few digital hiccups, but this is the only 'breaking' issue - everything else can be easily fixed by hand.

Thanks again!

[duak]

SethGI, The THS4121 has a much lower input bias current but it also has a minimum Common mode Input Voltage of 0.65 V (worst case - see p.4).  It could be made to work if the transducer was raised off DC GND by a volt or so - see the Analog Devices ap note.

[Henrik_V]

My comments:
You want to use a charge amplifier (CA) , because your source is a charge source, and the charge amplifier is a virtual short (ideal 0\$\Omega\$) , so connecting cable capacity influence is minimized. You want the inner conductor to be at the same potential than the shield and any charge produced by the piezo should not build up a voltage, instead it is sucked by the CA.
If you use a coaxial connection (SE) and connect the shield to GND (makes sense ) AND connect a charge amplifier , that amplifier absolutely needs a bipolar supply (or needs to be decoupled by an capacitor >>C_f ). 

SE to Diff in the second stage...

The value of R2 (noise source) is questionable ....  again, the input  impedance of a CA is 0\$\Omega\$ (ideal)  , in reality I measured values between 10 - 600\$\Omega\$ for lab grade commercial CAs.     A protection current limiting resistor (if needed) should be directly at the OP input (in the loop) ...
And stop thinking with voltages ... think about some fractions of pC charge  ... (isolation , guarding, cleaning ... ) until you hit the inverting input of the OP.
And remember C2 (and C1) are prone to vibration pickup ... (and other caps on the board can act as piezo actors ....)   

[SethGI]

Henrik,

Thanks for this, it's super helpful. I just want to clarify one thing on the schematic. What is that V12 doing? To me it looks like a voltage source (1.6V) shorted straight to ground, but I'm likely misreading it (on account of being a novice).

Also, what is Aol? Sorry, I'm a bit out of my depth here... thanks for all the help!

My comments:
You want to use a charge amplifier (CA) , because your source is a charge source, and the charge amplifier is a virtual short (ideal 0\$\Omega\$) , so connecting cable capacity influence is minimized. You want the inner conductor to be at the same potential than the shield and any charge produced by the piezo should not build up a voltage, instead it is sucked by the CA.
If you use a coaxial connection (SE) and connect the shield to GND (makes sense ) AND connect a charge amplifier , that amplifier absolutely needs a bipolar supply (or needs to be decoupled by an capacitor >>C_f ). 

SE to Diff in the second stage...

The value of R2 (noise source) is questionable ....  again, the input  impedance of a CA is 0\$\Omega\$ (ideal)  , in reality I measured values between 10 - 600\$\Omega\$ for lab grade commercial CAs.     A protection current limiting resistor (if needed) should be directly at the OP input (in the loop) ...
And stop thinking with voltages ... think about some fractions of pC charge  ... (isolation , guarding, cleaning ... ) until you hit the inverting input of the OP.
And remember C2 (and C1) are prone to vibration pickup ... (and other caps on the board can act as piezo actors ....)   

[ehughes]

Why are you connecting one side of the hydrophone to circuit common?     I think you might be affecting the DC condition on on side of the charge amplifier.
If you are doing fully differentlial,  the GND connection is not needed and may be a counter productive.

This circuit (ignore values) has worked well for me and I have used it with a piezo sensor and a CS5366 ADC.

[SethGI]

Why are you connecting one side of the hydrophone to circuit common?     I think you might be affecting the DC condition on on side of the charge amplifier.
If you are doing fully differentlial,  the GND connection is not needed and may be a counter productive.

This circuit (ignore values) has worked well for me and I have used it with a piezo sensor and a CS5366 ADC.

Yeah, you're totally right about that. We're getting ready to do a new version without the GND connection (just need to prototype first). Was that the only stage you had, or was there more? Just for my reference. Also, how was noise performance?

[Henrik_V]

What is that V12 doing? To me it looks like a voltage source (1.6V) shorted straight to ground, but I'm likely misreading it (on account of being a novice).

Also, what is Aol? Sorry, I'm a bit out of my depth here... thanks for all the help!

V12 is your Vcc/2 .. , but if you have your powersupply onboard why not have a bipolar supply for the input amp?
Aol : (DC) open loop gain of the OP

[Henrik_V]

Why are you connecting one side of the hydrophone to circuit common?     I think you might be affecting the DC condition on on side of the charge amplifier.
If you are doing fully differentlial,  the GND connection is not needed and may be a counter productive.

This circuit (ignore values) has worked well for me and I have used it with a piezo sensor and a CS5366 ADC.

Should work, but your cable runs at VMID potential, prone to cable microphony or triboelectric  effects. If your cable is moving/vibrating you quickly get unwanted charge signals ...  can be annoying, depends on the application.