Bambino, When you say grounded do you mean that the negative(-) of the accelerometer is connected to the accelerometer's shielding/housing? If so this could still be allowed to float with respect to your circuit boards Vss...did you ground the board's and/or accelerometer's Vss to the 20lb metal weight?...could you add some more detail. Tracy's circuit is really nifty and may help to resolve two problems at once.
If the case is not grounded to other metal in the structure, it still might be possible to float the case and to use the circuit.
If not, here is a variation that allows one side of the sensor to be grounded. You see, it takes more parts, but those could all be SOT23 or smaller. The voltage reference is 0.4 Volts, and the action of feedback holds the voltage across the 100 Ohm resistor also at 0.4 volts, so 4 mA flows in the collector-emitter circuit (a p-mosfet would also work instead of the PNP superbeta transistor). The optional second op-amp buffers the output to the ADC. Note the 7.2 volt power supply for extra head room.
The diode is indeed 2mA. I have it soldiered in next to the first one. I'll Have it in the lab in the morning and hopefully some results to post this weekend.
Well it started out good, The drops were dead on, tight grouping. but then.................
After about 6 drops as the device was warming up The trigger started malfunctioning and when I did get a good shot, off the results were off the charts. I've mulled it over in my head and I can understand higher voltages, but the max is measured from the current baseline to the peak. This would suggest > 2mV perG, which I don't understand.
I am about to open it up and check for physical damage, but I doubt I will find any as the device did perform repeatedly.
How's it going with the squiggles in the response? Maybe having the bias set at 4mA will help, and if that doesn't do it, then maybe see what happens connecting them directly to the 7.2V instead of the 5V. A little more voltage makes a stiffer current source.
Possible contribution of the capacitor to distortion of the peak came up earlier. Maybe. I've been looking at the ECPU series capacitors from Panasonic, "coupling circuit for audio". They are PPS film SMT capacitors, available up to 1 uF in a 1210 package. Low ESR, near zero voltage dependence or piezo effect, low soakage, and intrinsically non-polar.
The Squiggles did soften up a little, but still present. The big improvement was the accuracy of the peak. I can atleast calibrate now! Before, three drops would vary too much and I am satisfied it was because I was starving the sensor. O G's registers 2.646V now. Much better. The head was allready at the lab yesterday, so I had no way of charging it before the tests. The batteries are below the rated voltage right now so that may account for the non linear results I ran unto shortly after starting. I'll charge it up tonight and test it here in the morning. I don't have a calibrated rig here at my shop, but I do have a tube capable of answering this latest question.
Do you really think a 1uF is enough not to clip the signal. It just seems small, not that I have devine insight, or I wouldn't be asking.
The only thing that is loose is the 3" of wire suppling power to the board. I doubt it would be responseable for the blips. Plus note that when the malfunction toke place, the blips dissapeared! If I am right, and a fully charged battery gets rid of the malfunction, then the blips are an electrical by product. No epoxy used currently on the prototype.
The software filter basically just smooths out the signal, hard to describe an iir filter. I can say that it is not in an unrolled loop in my code, so the glitch would show up repeatedly as the loop processed the data.
Here is the engine that gives me the filter, thanks to some forum members back a few months ago. http://www-users.cs.york.ac.uk/~fisher/mkfilter/trad.html
The resulting code:
static void filterloop()
{ for (;[noparse];)[/noparse]
{ xv[noparse][[/noparse]0] = xv; xv = xv; xv = xv; xv = xv;
xv = next input value / GAIN;
yv[noparse][[/noparse]0] = yv; yv = yv; yv = yv; yv = yv;
yv = (xv[noparse][[/noparse]0] + xv) + 4 * (xv + xv) + 6 * xv
+ ( -0.6687309839 * yv[noparse][[/noparse]0]) + ( 2.9435650706 * yv)
+ ( -4.8735775075 * yv) + ( 3.5982821869 * yv);
next output value = yv;
}
}
I just convert it to Visual Basic since I am not very proficient in C
I did have physical damage after all. One of my through Hole soldiers had a nasty little spike on it. Over time this poked through the insulation coverering the bottom and was·touching the top of the sensor!
At low G's the squiggles are gone, and Higher G's do see a diffinate improvement with the 4mA upgrade!
Q1:
I've about finished redesigning·my board and thought I would get a sanity check on the op amp.
Since the Full Scale Voltage seen by the ADC is going to be plus or minus 1 volt change, then any slew rate above 1V/us should be sufficent, right?
Q2:
I said the circuit was sensitive to temperature change!
What is more prone to be the weak link?
1.Propeller clock (crystal changeing the rate at which I'm driving the ADC.)
2.ADC(Sample and hold circuitry)
3.Current diode(jfet)
4.De-coupling capacitor.
5.Accelerometer
This would help me place my temp sensor more strategically.
Q3:
The bottom plane of my board is connected to ground. Would placeing a plane around the sensor area on the top of the board help with the heat displacement,
and if so, should I connect it to ground, nothing, power....etc
Q1: 1 volt, yes, but what is the time scale? I scanned back at the time course results you had posted, but no units for X and Y. My guess is that a 1V/µS op-amp would be fine. Maybe even slower. This will also depend on how much voltage gain you need from the op-amp. Have you changed your original circuit? As I recall, you were constrained by a design already cast in copper.
Q2: The current diode is the first thing I would look at, especially if it is being operated with as low a voltage as previously discussed. That is a weak link. I don't know anything about the accelerometer. If you have a extra diode, you can measure its temperature dependence. Build a better current source?
Q3: Implying that something is getting hot? On 4 mA?
Q1: The whole window is about 22mS, The actual bell curve of the drop is ~11ms
Q2: The diode is now 2 diodes in parallel feed by 7.2 to 8.4 volts, bypassing the 5 V regulator
Q3: The unit does build up heat when I am running the Converter, but seems to run a little cooler when waiting for a drop command. When waiting the Chip select on the ADC is not selected.· There is not really a heat problem, Maybe 5 to 10 celcius above ambient, but was thinking that a plane covering this area and touching the temp sensor would give me closer to real time what the temperature is around the hot spot.
Not hearing from anyone I choose an op amp capable of video slew rates and put it in the design. It's overkill, but does have a nice power down feature I liked.
"As I recall, you were constrained by a design already cast in copper"
Yes the prototype I have is about as post rigged as it can get. But those changes and these concerning the op amp are at the moment still in Gerber files.
I have sent it off for a quote, but not too late to change. The things cast in stone are the amount of space I have to work with. 2.5 inch circle is all I got, and the op amp, plus temp sensors left me barely enough room for the fuducials.
Wow--130 MHz bandwidth and 375 V/µS slew rate. That should cover the speed issue. Your original circuit doesn't have an op-amp. Is it going to be a gain=1 buffer between the signal and the ADC? The only thing that jumps out at me is the limited common mode range at the input (up to 2.8 volts) if used on a 5 volt supply.
Be sure the board layout is tight with good bypass capacitors. These high speed op-amps are very happy to break into oscillation given parasitic feedback.
Yes, It's a straight voltage follower. I have the -In pin connected to out with a short trace under the chip. All chips are de-coupled except for the op amp its self. The out pin has a very short trace to the ADC channel pin.
The capacitor feeding the amp is the recommended one from dytran for decoupling the signal from the sensor.
A Tanalytic cap is what they call it!
I've decided to try one of the circuits described by Tracy. Which one is now the question.
It depended on whether or not the sensor floated, so I'll describe the circuit.
The power, as discussed, flows through the cc diodes into the sensor, the ground wire coming back from the sensor attaches to ground for the circuit board. The circuit board is encased in PVC and does not touch the stainless steel. However the data sheet for the sensor says that the sensor houseing is connected to ground internally. The sensor housing is connected to the stainless steel.
So which circuit should I be trying to use?
FYI. The circuit changes worked great in regards to getting rid of the glitch seen on the charts it produced. The decission to try one of the mentioned circuits is based on the fact that the readings still have to account for temperature.
It's too bad you can't float the sensor, because that would be the simpler circuit
If one side of the sensor is grounded, then this one is a very serviceable current source: . It can probably run off of the 5V regulated supply if the components are all low-voltage capable, and then the 0.4V reference can be a simple voltage divider. The transistor should be a logic-level p-mosfet or a superbeta pnp, as shown. You already have the buffer in place after the sensor, the ADA4851. I don't think you need anywhere near the bandwidth of the AA4851, but I don't remember how fast a process this is. That will also have something to do with how to choose the other components.
You are right about the bandwidth. Not the greatest on calculating such things, so I shot for overkill. The boards came back in good shape and worked great. I am trying to get a design change through that would allow the sensor to float, but heads have allready been made in this manner, so I will try the second method. Maybe put it on a riser board as an add on to the existing boards. The last set was such a success that there are now 4 prototypes instead of one, so being able to retro-fit them would be great.
I am sampling the signal at about 57k. The drop from start to finish is ~11ms, and my recorder is taking in 2k samples. I send the readings to a buffer and monitor the buffer from another cog for a trigger condition. Once detected it shuts down the ADC circuit and transmits the data.
For a voltage divider to keep the 0.4 reference voltage I couldn't keep the power from the battery source, it would have to come from the regulated 5 volts. Any ideas on how to keep the higher voltage.
Could the circuit work as drawn except have just the reference come from 5 volts and the rest be powered via the 7.2 to 8.4 battery supply?
Well, that circuit as you have drawn it won't work. The feedback would operate to keep 0.4 + (7.2-5) volts across the 100Ω resistor, and that is not what you want. The 0.4V and the 100Ω resistor have to reference the same node.
Here is a circuit that uses only the 5V regulated supply, and the 0.4V reference is made with a simple voltage divider. The op-amp will have to be a single supply type that has an common mode range that extends up to close to the +5V rail, and good accuracy, but there are many to choose from these days. It does not need extraordinary bandwidth. I suppose I would start with an LT1783 in SOT23, simply because it is one I keep on hand. The transistor could also be SOT23 or smaller. If you did not already have the buffer amplifier installed on the PCB, you could use an dual op amp for both the current source and the buffer.
Comments
No, I don't ground to the metal.
If not, here is a variation that allows one side of the sensor to be grounded. You see, it takes more parts, but those could all be SOT23 or smaller. The voltage reference is 0.4 Volts, and the action of feedback holds the voltage across the 100 Ohm resistor also at 0.4 volts, so 4 mA flows in the collector-emitter circuit (a p-mosfet would also work instead of the PNP superbeta transistor). The optional second op-amp buffers the output to the ADC. Note the 7.2 volt power supply for extra head room.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Tracy Allen
www.emesystems.com
Post Edited (Tracy Allen) : 2/17/2010 6:53:38 PM GMT
I'll have busy weekend for sure.
After about 6 drops as the device was warming up The trigger started malfunctioning and when I did get a good shot, off the results were off the charts. I've mulled it over in my head and I can understand higher voltages, but the max is measured from the current baseline to the peak. This would suggest > 2mV perG, which I don't understand.
I am about to open it up and check for physical damage, but I doubt I will find any as the device did perform repeatedly.
Possible contribution of the capacitor to distortion of the peak came up earlier. Maybe. I've been looking at the ECPU series capacitors from Panasonic, "coupling circuit for audio". They are PPS film SMT capacitors, available up to 1 uF in a 1210 package. Low ESR, near zero voltage dependence or piezo effect, low soakage, and intrinsically non-polar.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Tracy Allen
www.emesystems.com
Do you really think a 1uF is enough not to clip the signal. It just seems small, not that I have devine insight, or I wouldn't be asking.
And how does your software filter change the signal?
Graham
The software filter basically just smooths out the signal, hard to describe an iir filter. I can say that it is not in an unrolled loop in my code, so the glitch would show up repeatedly as the loop processed the data.
Here is the engine that gives me the filter, thanks to some forum members back a few months ago.
http://www-users.cs.york.ac.uk/~fisher/mkfilter/trad.html
The resulting code:
static void filterloop()
{ for (;[noparse];)[/noparse]
{ xv[noparse][[/noparse]0] = xv; xv = xv; xv = xv; xv = xv;
xv = next input value / GAIN;
yv[noparse][[/noparse]0] = yv; yv = yv; yv = yv; yv = yv;
yv = (xv[noparse][[/noparse]0] + xv) + 4 * (xv + xv) + 6 * xv
+ ( -0.6687309839 * yv[noparse][[/noparse]0]) + ( 2.9435650706 * yv)
+ ( -4.8735775075 * yv) + ( 3.5982821869 * yv);
next output value = yv;
}
}
I just convert it to Visual Basic since I am not very proficient in C
At low G's the squiggles are gone, and Higher G's do see a diffinate improvement with the 4mA upgrade!
Glad you are getting this sorted!
Graham
I've about finished redesigning·my board and thought I would get a sanity check on the op amp.
Since the Full Scale Voltage seen by the ADC is going to be plus or minus 1 volt change, then any slew rate above 1V/us should be sufficent, right?
Q2:
I said the circuit was sensitive to temperature change!
What is more prone to be the weak link?
1.Propeller clock (crystal changeing the rate at which I'm driving the ADC.)
2.ADC(Sample and hold circuitry)
3.Current diode(jfet)
4.De-coupling capacitor.
5.Accelerometer
This would help me place my temp sensor more strategically.
Q3:
The bottom plane of my board is connected to ground. Would placeing a plane around the sensor area on the top of the board help with the heat displacement,
and if so, should I connect it to ground, nothing, power....etc
Thanks.
Q2: The current diode is the first thing I would look at, especially if it is being operated with as low a voltage as previously discussed. That is a weak link. I don't know anything about the accelerometer. If you have a extra diode, you can measure its temperature dependence. Build a better current source?
Q3: Implying that something is getting hot? On 4 mA?
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Tracy Allen
www.emesystems.com
Q2: The diode is now 2 diodes in parallel feed by 7.2 to 8.4 volts, bypassing the 5 V regulator
Q3: The unit does build up heat when I am running the Converter, but seems to run a little cooler when waiting for a drop command. When waiting the Chip select on the ADC is not selected.· There is not really a heat problem, Maybe 5 to 10 celcius above ambient, but was thinking that a plane covering this area and touching the temp sensor would give me closer to real time what the temperature is around the hot spot.
Post Edited (bambino) : 3/21/2010 11:52:26 PM GMT
Yes the prototype I have is about as post rigged as it can get. But those changes and these concerning the op amp are at the moment still in Gerber files.
I have sent it off for a quote, but not too late to change. The things cast in stone are the amount of space I have to work with. 2.5 inch circle is all I got, and the op amp, plus temp sensors left me barely enough room for the fuducials.
Be sure the board layout is tight with good bypass capacitors. These high speed op-amps are very happy to break into oscillation given parasitic feedback.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Tracy Allen
www.emesystems.com
The capacitor feeding the amp is the recommended one from dytran for decoupling the signal from the sensor.
A Tanalytic cap is what they call it!
It depended on whether or not the sensor floated, so I'll describe the circuit.
The power, as discussed, flows through the cc diodes into the sensor, the ground wire coming back from the sensor attaches to ground for the circuit board. The circuit board is encased in PVC and does not touch the stainless steel. However the data sheet for the sensor says that the sensor houseing is connected to ground internally. The sensor housing is connected to the stainless steel.
So which circuit should I be trying to use?
FYI. The circuit changes worked great in regards to getting rid of the glitch seen on the charts it produced. The decission to try one of the mentioned circuits is based on the fact that the readings still have to account for temperature.
If one side of the sensor is grounded, then this one is a very serviceable current source:
Could the circuit work as drawn except have just the reference come from 5 volts and the rest be powered via the 7.2 to 8.4 battery supply?
Here is a circuit that uses only the 5V regulated supply, and the 0.4V reference is made with a simple voltage divider. The op-amp will have to be a single supply type that has an common mode range that extends up to close to the +5V rail, and good accuracy, but there are many to choose from these days. It does not need extraordinary bandwidth. I suppose I would start with an LT1783 in SOT23, simply because it is one I keep on hand. The transistor could also be SOT23 or smaller. If you did not already have the buffer amplifier installed on the PCB, you could use an dual op amp for both the current source and the buffer.