Odd 9V Power Drain

[spamiam]

So last night I discarded the regulator and started with a new LM2940T and put together a circuit board with a very large heat-sink, a 1000uF capacitor for VIN, then a 1000uF for VOUT, and another 10nF filter on the power rail of the breadboard.

The ESR of your capacitance on Vout may still be too high. Read the datasheet on the LM2940. They are quite specific about the ESR.

The output capacitor is critical to maintaining regulator stability,
and must meet the required conditions for both ESR
(Equivalent Series Resistance) and minimum amount of capacitance.
MINIMUM CAPACITANCE:
The minimum output capacitance required to maintain stability
is 22 μF (this value may be increased without limit). Larger
values of output capacitance will give improved transient response.
ESR LIMITS:
The ESR of the output capacitor will cause loop instability if it
is too high or too low. The acceptable range of ESR plotted
versus load current is shown in the graph below. It is essential
that the output capacitor meet these requirements, or
oscillations can result.

http://www.national.com/ds/LM/LM2940.pdf

At low currents, the ESR needs to be very close to 0.5 ohms, and no more than 1 ohm. At higher currents you can get up to 5 ohms. I'd bet you are drawing less than 200mA and therefore you need no more than 1 ohm. It is hard to determine the ESR of most general purpose electrolytics. It is often not specified in the data sheets. Some of them DO have sufficiently low ESR, though.

-Tony

[liam.zbasic]

Did some research on ESR. This parameter is a moving target that changes with usage and time. An ESR meter seems like a good investment. For the sake of argument, what happens if the equivalent resistance is 5 ohms? I read that this could lead to a regulator instability, but what does that physically mean? More fried wires? Right now, the GBOT seems to be functioning fine with my new regulator circuit with the large heat sink and 1,000uF (25V) filters on VIN and VOUT. In fact, I was able to remove my other 1,000uF filter on the power supply of the IR range sensors. The IR readings are stable. I'll watch out for suspicious behavior. Thanks for your tips.

[GTBecker]

Did you measure the old regulator input resistance before you discarded it?

[liam.zbasic]

I did not measure the input resistance of the old regulator. I do not own an ESR meter. Even if the input resistance was acceptable, that does not guarantee its integrity. It was involved in a fire, so its out. Safety first. Anyone know where I can find a 1000uF low-ESR DIP filter? Always appreciate your comments. Thanks.

[GTBecker]

> ... I do not own an ESR meter...

A simple ohmmeter would be sufficient to find a short. While the
localization of the damage is a good indicator, it is not convincing
evidence that the regulator failed. I believe I have never had a
regulator develop an input short - although I've surely cooked some
number by overload - and haven't needed to concern myself with ESRs to date.

You had an unusual, dangerous failure. It would be most safe to know
the cause.

[spamiam]

Anyone know where I can find a 1000uF low-ESR DIP filter? Always appreciate your comments. Thanks.

I am sure that there is a way to measure ESR, but I certainly don't know. I presume it would involve an oscillator and the cap under test, and measuring current through a known precision resistor. IFAIK, ESR is essientially the "internal resistance" of a capacitor. An ideal capacitor has none. I think the ESR would be the extra reduction in current flow compared to an ideal RC circuit. I think it would be hard to measure without precision equipment.

Most of the datasheets for the run of the mill electrolytic caps do not specify ESR. Maybe the ESR can be extraploated from other data which is included in the datasheet, but it is beyond me. Some caps DO have the ESR specified. Usually they are listed as "Low ESR" types. Presumably the rest of the caps have higher ESR, but without a specification, I can't tell you how much higher!

The datasheet for the 2940 regulator makes special note of the ESR, which I take to mean that many run of the mill caps would NOT meet the requirements, but I just don't have the data to be sure one way or the other.. I solved this by getting a low ESR tantalum cap for my application because I was also quite concerned about the temperature coefficient to the ESR. The regulator might well see rather low or high temps (automotive).

I guess that if you have an oscilloscope, you can look at the output and see if there is a high frequency oscillation superimposed on the DC.

There are several 1000uF electrolytic caps listed in Digikey with sufficiently low ESR at room temp. This is one: SEK102M016ST. Maybe most electrolytics do have sufficiently low ESR at room temp, but I can't prove it.

-Tony

[liam.zbasic]

I'm currently using the 493-1065-ND (Nichicon 1000uF 25V, VR-series) from Sparkfun.com. Digikey and the Nichicon datasheets do not spec out ESR values. This is suspicious. I'll check out the SEK102M016ST capacitor you recommended - the datasheet lists 0.24ohms. Less IR sensor noise is always better. Especially for my next project.

[dkinzer]

[T]he Nichicon datasheets do not spec out ESR values. This is suspicious.

Any capacitor being marketed as "low ESR" will almost certainly have an ESR specification. If it doesn't have that specification, you can be pretty certain that it does not fall into the category of "low ESR". Consequently, when selecting a capacitor for a low ESR application, you'll need to restrict your search to those touting "low ESR".

[liam.zbasic]

Okay, I replaced all the 1000uF capacitors with low ESR versions (SEK102M016ST). There are filters are on the VIN and VOUT of the voltage regulator, and also on the power terminals to the IR range sensors very near the sensors. All is behaving normally. No fires. Also added a 3.0V voltage reference in the form of a clamp diode. This improved resolution results in smoother GBOT response. There's just one nagging issue. If I disconnet the power supply to the motors and fix the GBOT and target and record IR distance time response, I still have some noise (0.05" amplitude, 0.1" peak-to-peak). Its signficantly lower than before, but there seems to be a ~1Hz LCO lingering (plot below). I also changed out the IR sensors with new ones. No change. Could it be that I have to many filters on the +5V power bus? Or is this as good as it gets with Sharp IR sensors? Thanks.

[spamiam]

Okay, I replaced all the 1000uF capacitors with low ESR versions (SEK102M016ST). I still have some noise (0.05" amplitude, 0.1" peak-to-peak). Its signficantly lower than before, but there seems to be a ~1Hz LCO lingering (plot below). I also changed out the IR sensors with new ones. No change. Could it be that I have to many filters on the +5V power bus? Or is this as good as it gets with Sharp IR sensors? Thanks.

The noise is lower with the low esr caps than with the previous ones?

I agree, here does seem to be a 1Hz oscillation in the signal. I'd call it an 'oscillation' rather than 'noise' (I always consider, perhaps wrongly, that 'noise' is fairly random). What does the 5V rail look like? How about the voltage feeding the 5V regulator? How about the VCC to the sensor(s)?Is the sensor signal you showed after the butterworth filter? If so, what does the signal look like before the butterworth?

It would be interesting to switch one of the large caps (to a value half and double) at a time to see if the period of the oscillation changes. something there may be prone to oscillation, and may be better damped with different capacitances. What sort of oscilloscope are you using? I have a decent (old) one without storage, and I would love to have some sort of storage function!

-Tony

[liam.zbasic]

The noise improved with low ESR filters (~0.1 inch amplitude to ~0.05 inch).

I do not own an oscilloscope or a logic analyzer. I have a voltmeter. The previous response plot is based on "debug.print" output copied to Excel and includes the digital filter. The debug command slows execution speed from 120Hz to 25Hz on a ZX-40a device. So the ~1Hz oscillation could be an aliased mode. The digital filter in this case is an 8th order "running average" since the butterworth coefficients are designed for the 120Hz program (i.e., no "debug.print").

Without the digital filter, the noise grows from 0.05" amplitude to 0.25". Without the 1000uF filter near the IR sensor power pins, the amplitude grows another 0.25" (I'll have to re-verify that, could be more).

To this point, the noise has been reduced dramatically. I'll try different capacitors, which is difficult at this point since all the filters are soldered. But I'll give it one last ditch effort because I need the cleanest signal possible for the gbot90.

Thanks for your tips.

[spamiam]

The noise improved with low ESR filters (~0.1 inch amplitude to ~0.05 inch).

I'll try different capacitors, which is difficult at this point since all the filters are soldered. But I'll give it one last ditch effort because I need the cleanest signal possible for the gbot90.

Does the datasheet for the sensors describe the variability of the readings under optimal conditions? 0.05" variability seems pretty good to me.

You could mock up the sensor and filters and power supply on a breadboard to check the performance of the circuit.

It seems that you may have reached the point where you need some sort of oscilloscope. For this you don't need a high performance 200MHZ digital storage scope. I think you can make a decent PC based oscilloscope using your sound card. There are a bunch of options to be found on the internet.

-Tony

[liam.zbasic]

Yes, I'm probably hitting the limits of the GP2D12 IR sensor. The datasheet does not discuss variability. See link below:

http://media.digikey.com/pdf/Data%20She ... 060207.pdf

Last night I changed my target to a smooth white background. I kept the low-ESR capacitors in the GBOT, and disconnected the motors. I set the target ~10 inches from the GBOT. I recorded "debug.print" readings with and without the digital filter and plotted the results (see attachments). The unfiltered signal modulates about the mean with ~0.1" amplitude. Some extreme outliers observed (0.2" amplitude). The filtered signal is 0.025" with no extreme outliers. Previous replies and videos are with respect to a tan cardboard background and noise was a bit higher (0.05"). I'll plot capacitor effects later this week.

[spamiam]

Yes, I'm probably hitting the limits of the GP2D12 IR sensor. The datasheet does not discuss variability.

I checked the datasheet. As you say, they do not show error bars on the distance measurements. I am not sure what is the intended application for the sharp sensor. Clearly it is for distances greater than 10cm or so due to the ambiguity of readings at short distances. It may be that precision is not that important for the intended application.


It might be useful for you to determine the "noise" level in the readings because that would probably equate to the dead-zone in your control.

I was considering the issue of "integral wind-up". There are two PID equations. One is "Position" and the other is "Velocity". I believe it is the "Velocity" equations that are not subject to wind-up. It essentially uses proportional, first derivative and second derivative of the "velocity". I forget the details of the equations, but you might be interested in this approach. As a matter of fact, I have to re-look at this myself.

-Tony

[liam.zbasic]

I went ahead and rotated the GBOT 90 degrees to balance on its 2 wheels. I disabled the IR range sensors and added a new GP2D120 IR sensor. The result is the GBOT90 shown in the link below. Noise and resolution was an issue but manageable. The Vref was set to 3.0V. No gyros and no acceleromenters were used, just the single GP2D120 for pitch rate & angle measurements. I used the same PID control system from the horizontal GBOT with updated gains. Proportional and integral gains were much higher. Integral gain was highest.

http://www.youtube.com/watch?v=bxx14Xe2iNg