LM2907-N: LM2907 F to V internal servo loop appears to be underdampsed

Just to give you the degree of overshoot - it's about 1% and took a full 45 minutest to get back to the real value. Not much I know, but throws doubt on our gear.

Hi Jim,

You mention "While prototyping this I notice the output DC voltage appears to hunt after an input frequency step change. It exhibits the classic behavior of an underdamped servo loop whereby the initial change in the input F creates an overshoot on the output in V. Thereafter this rings out. I went back to observe the original single ended design  - does the same thing." The LM2907 is not a sophisticated device and output ripple is normal. As the datasheet mentions on page 10, "C2 acts as a filter to smooth the pulses of current and does not affect the output voltage. However, the size of C2 determines both the output response time for changes in frequency and the amount of output voltage ripple." Therefore, there is only so much that can be done with C2 with regards to the output waveform. If we can see what the output looks like there may be something that can be done to condition the waveform.

Would it be possible to see your LM2907 circuit, a description of the input stimulus, and the signal at the emitter output? It is difficult for us to assess what you are observing without a bit more information.

Regards, Thomas

Precision Amplifiers Applications Engineering

Here ya go.

Jim

Hi Jim,

I expect with the correct filter at the LM2907-N output you would be able to improve the output voltage ripple and response characteristics. TI produces the LM331, which is described as a Precision Voltage-to-Frequency Converter, but it may also be applied as a Precision Frequency-to- Voltage Converter.

The LM331 has a good applications note that shows how to add a filter to improve the ripple and damping: 

AN-C V/F Converter ICs Handle Frequency-to-Voltage Needs (Rev. B)

This should be applicable to the LM2907-N, or you could move to the LM331. Here's a link to its datasheet:

https://www.ti.com/lit/ds/symlink/lm231.pdf

Regards, Thomas

Precision Amplifiers Applications Engineering

Thanks for the info....but I have now ordered and recieved the incorrect variant of the LM2907/17 twice. It has delayed my development project incredibly.
For the life of me I can't get out of the spec sheet what the part number is for the variant that is a 14 pin SOIC with differential input and no zener.
Above is a representation of what I want from the spec sheet.

What is the correct part number?

Interesting - my spec sheet equivalent, page 2, does not label the diagrams with part numbers, rather it labels them with package numbers and, unfortuantely,  these numbers never appear again in the entire document.

I must have an old rev of spec sheet? 

I'm still working on my original issue - slow response vs ripple. i plan to implement the butterworth filter. Let ya know.

Hi Jim,

We do try to keep things up to date on TI.com and the latest product datasheet is usually provided on a product's TI webpage. There are other sources on line that provide TI datasheets, but we have no control over them and they have older revisions posted.

Do let us know the results you obtain from the LM2907 after the Butterworth active filter is added at the output.

Regards, Thomas

Precision Amplifiers Applications Engineering

Meant to say . . . they "may" have older revisions posted.

Thomas

I did order the correct device after all - LM2907M.  I Had a different problem related to an incorrect assembly.

However, the 2907M it exhibits the same problem the LM2907-8 does. Always overshoots the final DC value by 1.5% then eventually zeroes in. If I decrease the ripple capacitor value it settles quicker but still overshoots by the same amount.

Am I going to get there with this device?

I can't tolerate that overshoot.

BTW, I am monitoring a variable reluctance sensor with a max F of 200hz.

Is that too low a freq for this device?

I don't think I'll bother with the butterworth filter - pretty sure it won't correct RMS overshoot.

Does the product engineer know about this overshoot?

Hi Jim,

In response to your LM2907 questions:

However, the 2907M it exhibits the same problem the LM2907-8 does. Always overshoots the final DC value by 1.5% then eventually zeroes in. If I decrease the ripple capacitor value it settles quicker but still overshoots by the same amount.

Am I going to get there with this device?

Based on your earlier results, you won't get the level of performance you are needing with the basic LM2907 alone. That is why we suggested adding the active filter at the output.

I can't tolerate that overshoot.

The active filter should help. A Butterworth response has a Q of 0.707 and is optimally damped.

BTW, I am monitoring a variable reluctance sensor with a max F of 200hz. Is that too low a freq for this device?

The LM2907 shouldn't have any issue functioning as intended at 200 Hz, or less. Many of the automotive sensors that it was originally designed for had applications that crossed this frequency range.

Regards, Thomas

Precision Amplifiers Applications Engineering

I let it run overnight - the output continues to drop.
Is this due to self heating?
I have a 10k load on the transistor emitter output. If raised this by 10x, ie 100k, would the heating be reduced?
Let's see i have the non-zener, differential variant set up powered by 10.2 volts with precision regulator. This provides about an 8 volt output span. I've selected the RC to max out at 200hz.After 14hrs of operation at the same frequency input, 100hz, ( I check it - no movement), the output has dropped about 2% of 8 volts or 160mv. That's 2.8% of it's initial output of 5.73v. Granted most of that drop occurs in the first 15mins and perhaps that is overshoot which the filter might fix I suppose. Guess I'll try it. I see graphs for normalized output over temperature change. What I don't see is a "long term drift" graph. - Should be included.

I graphed my data - most of the drop occurs in the first 100mins then levels out, however doesn't stop dropping. Wow, to filter this out I would need a corner frequency of 170 micro hertz - impractical. This isn't overshoot, this is drift so I don't think I'll implement a filter with a huge capacitor/resistor after all.

I heated up the board with a heat gun. I learned something - the slightest amount of ambient temperature change decreases the output by 10s of millivolts.

Here's my current hypothesis. The charge pump capacitor is changing value with temperature which is causing the output to change proportionally. I used a cap with an X7r temperature coefficient which is generally quite good but even that can change by 12% over its' temperature range. So I'm going to try a COG type ceramic that preports to be the most temperature stable cap you can get.

I see a slow, 1 milivolt every 3 second, drop in the V on the output if i put in a constant well regulated frequency. - I'll bet the cap/assembly is self heating?

Hi Jim,

Overlooking capacitor quality is a common issue in timer, sample-hold, and V-to-F/ F-to-V circuits. X7R capacitors have a high, non-linear temperature coefficient in comparison to a capacitor having a quality dielectric/composition. I would never apply them in a critical timing, sampling, or filter applications. They are fine for simple power supply bypassing and ac coupling applications where the capacitance over temperature and voltage is not critical.

The C0G dielectric is formulated for very good temperature stability and is recommend for the LM2907 C1 and C2 capacitors. C2 is a higher capacitance than C1 and larger values may be more difficult to obtain in C0G. In such cases, film capacitors having a low temperature coefficient dielectric should be sufficient. It is important to use low temperature coefficient resistors in the circuit as well.

The LM2907 is a general purpose F-to-V, V-to-F. You may find that the LM331, which is specified as a precision voltage-to-frequency converter, provides more precise performance compared to the LM2907. If you haven't considered it for your F-to-V application do have a look at the datasheet Figure 19, Precision Frequency-to-Voltage Converter. Observe that there is a note associated with Figure 19, " *Use stable components with low temperature coefficients."

www.ti.com/.../lm231.pdf

There is the LM331 Applications Note that I cited earlier as well.

Thomas

Precision Amplifiers Applications Engineering

I tried the COG capacitor and this pretty much killed the drift, ie less than 0.2% overshoot typically.

Onto the next problem - The reason I switched from the LM2908 to the LM2907 is that I wanted a differential input so that common mode offsets and noise would be rejected. Experiments have shown the differential device does not reject any common mode signal AC, DC, or otherwise whatsoever. Wow, run that one up the flag pole.

A second experiment I performed was to use an opamp configured classic differential in front of the 2908 single ended variant - this works wonderfully to reject common mode signal.

I was hoping to reduce the circuity due to space constraints  by using the diff variant. I guess not.

What gives?