Experimental Notes of 12/20/2002

This account gives details of some of the modifications which I have tried on my Type LX1358 LVDT sensor, marketed as a Kit by Nuova Elettroncia. By experiment, I have found it possible to significantly lower the circuit noise and I have also been able to increase the time constant of the output from 1 second up to 47 seconds. This information, which I believe to be correct, is supplied as experimental notes in good faith but without any warranty or liability, either direct or implied. The suggestions are the result of my personal experiments and I have not discussed the circuit changes with Nuova Elettronica. I have no connection with Nuova Elettronica other than as a customer. The suggestions should not in any way be interpreted as a criticism of Nuova Elettronica equipment or designers.

Nuova Elettronica is an Italian Company which produce a wide range of Electronic Kits. Details of the Kits are published in their monthly Magazine. They marketed a 'garden gate' seismometer Kit which appeared in Issues 130-131 in May-June 1989. This featured the LVDT sensor Type LX922 and a microprocessor monitoring system with a thermal printer. A 1.4 second vertical pendulum seismometer Kit was described in Issue 195 of June-July 1998 and is still available. This featured the replacement LVDT Type LX1358 sensor, with an upgraded computer and printer system.

A Linear Variable Differential Transformer sensor is an excellent technical choice since it guarantees the linearity of response and also a very high sensitivity for a wide range of displacements between the seismic mass and the ground. The LX1358 sensor was originally designed to interface to a 12 bit A/D converter and some modifications are required to get optimum performance when it is used with a 16 bit A/D converter. The changes described relate to improvements in the stability, a reduction in the circuit noise, narrower band filters to exclude environmental noise and the use of integrated circuits with improved specifications.

I built the LX1358 and then used it unmodified on an NE 1.4 sec damped pendulum. LVDTckt1 The circuit is designed to drive a 0 to +10 V 12 bit A/D converter, with appropriate noise levels. Most current computer logging systems, like mine, use +/-10 V input 16 bit A/D Converters. The x8 increased resolution showed up significant circuit noise. The on board regulated power line is ~12 V DC and the signal zero reference line is approximately half this. This signal reference is set by two resistors in the NE5521 chip and the line is buffered by a TL081 opamp. The output opamp is a CA3130.

Modification 1

The final CA3130 opamp will drive the output from line to line, but it is an excessively noisy device. If you do need the line to line output drive capability, a TLC2201, while a bit more expensive, is vastly preferable. If you are re-referencing the output to 0 V and maybe adding additional gain and filtering, an OP07 works well. A TL071 or a LF411 will also work here, but their drift and noise are inferior to an OP07.

Modification 2

A more thorough study of the output noise revealed that the TL081 reference line buffer amplifier contained measurable amounts of 100 Hz hum as well as quite a lot of hash noise. This was feeding through the output opamp with the signal, since the +ve input line of the CA3130 was filtered by 4.7 K resistor and a 10 mu F capacitor. This opamp was changed to an OP07. The capacitor from earth to the centre of the resistor chain was changed from a 4.7 mu F electrolytic to a 47 mu F low leakage electrolytic and a 47 mu F Tantalum was connected from the opamp output to earth.

Modification 3

I have a site with a lot of environmental noise and passing trucks showed up vibrantly on the recordings! Since I had no wish to monitor road traffic, I decided to investigate the built in filter characteristics of the Sensor. I had some difficulty in understanding the frequency design of the phase detector circuit, so I measured the response. The readings were 3 dB down at 33 Hz, 6 dB down at 49 Hz and 20 dB down at 254 Hz. This is not too good for seismic use and will allow environmental noise (including cars and lorries) to be detected with ease. Environmental noise may be serious at frequencies above about 20 Hz. NE seem to have copied the Phillips data sheet circuit which was designed for commercial LVDT applications, but have not altered the filter. I changed three resistors and a capacitor to improve the noise rejection above 10 Hz and to give a three pole response with a much sharper cut-off. I measured the new 3 dB point at 10 Hz and the 20 dB point at a bit under 20 Hz. The cut-off frequency may be lowered still further if desired, to below 3 Hz, but two additional capacitor changes are needed. Miniature polyester capacitors with a 5 mm pin spacing are used. I measured the output impedance of the phase sensitive detector, pin 5, at 1.8 Ohm, so it does not effect the value of the input filter resistor R3. It is a voltage in / voltage out, direct / inverted, switched unity gain circuit.

There are three high frequency limiting circuits. The two pole active filter with C5 and C6 is the first, the capacitor C12 is connected in parallel with the feedback resistor R11 on the output amplifier and there is the R12 / C15 filter on the final output. R11 / C12 has a design rolloff frequency at 3.4 Hz. The final filter R12 / C15 has a rolloff of about 1,600 Hz. I am puzzled by this high value. It may have originally been provided to ensure the stability of the CA3130 from the effect of output cable capacity, but it did not prevent some noise aliasing from effecting my 16 bit A/D converter. I decided to re-calculate the overall response as a 3 pole filter and to use R11 / C12 as necessary to balance the rolloff of the high pass filter / differentiator circuit. The third low pass pole, R12 / C15, is applied across the output where wide band + 100 Hz noise on the reference line is more serious.

Considering the three pole active filter, you have the choice of-

a) keeping the same two pole filter capacitors C5 and C6 or

b) using increased capacities to lower the noise and / or the frequency rolloff.

It is desirable to limit the resistor values used in the filter circuit since the auxiliary opamp has a nominal input bias current of 0.21 micro Amp, max 0.6 micro A and a max. offset current of 50 nA. eg A 44 K Ohm input gives a typical offset of 9.24 mV, which appears on the output as a 32.8 mV error.

I calculated the following 10 Hz cut-off filters.

a) R3=24K, R6=20K, C5=470 nF polyester (supplied), C6=1 muF polyester (supplied), R12=1.5K, C15=10 muF solid Al electrolytic, C12=1.5 nF polyester

Rather surprisingly, I did not see an internal noise increase over the standard design with these larger value input resistors. The decrease in bandwidth seems to compensate for any increase in resistor or input current noise. No figures are available for the noise level of the auxiliary amplifier in the NE5521N chip. The electrolytic capacitor C15 should be measured / selected. Solid Al capacitors are preferable to a wet Al electrolytic. It may be necessary to trim the value of R12 to provide the 0.015 sec time constant, since the capacitor has a 20% tolerance. You will also need to recalculate R12 if a resistive output attenuator is used.

b) R3=4.4K, R4=22 K, R5=56K, R6=7.9K, C5=2.2 muF polyester, C6=3.3 muF polyester, R12=1.6K, C15=10 muF solid Al electrolytic, C1=1.5 nF polyester

This may be a better design. 3.3 mu F is the largest capacity that you can buy in miniature polyester capacitors with 5 mm pin spacing. (The components should be of appropriate size for the board, if at all possible.)

The resistor values may also be increased to give a cut-off frequency just below 3 Hz without running into noise / drift problems.

c) R3=15K, R4=55 K (=33 K + 22 K), R5=140 K (=130 K + 10 K), R6=27K, C5=2.2 muF polyester, C6=3.3 muF polyester, R12=5.4K, C15=10 muF solid Al electrolytic, C1=4.7 nF polyester. The increased values of the gain resistors R4, R5 are to match the input impedances to both filter amplifier inputs. This reduces the offset voltage caused by the rather high input currents.

Modification 4

The output of the two pole filter, pin 1 on IC1, is shown connected to the 22 mu F electrolytic condenser C7 and a 47 K Ohm resistor R9 into the inverting input of the output amplifier IC3. This choice seems to have been made for the original LX922 sensor when a 0 to +5V 8 bit A/D converter was used. The zero reference level used was then +2.5 V, so the electrolytic capacitor was correctly biassed. With the change in reference level to +6V for the LX1358, this capacitor may be reverse biassed by up to 5V.

The feedback resistor on IC3 is 10 Meg Ohm. An electrolytic capacitor has a typical leakage current of 0.01xCxV. For 5V and 22 mu F, this is ~1.1 micro A, which would give an 11 V leakage error on the output! I noted a very considerable offset and increased noise. I did not consider this to be satisfactory and I replaced C7 with a polyester capacitor.

You can buy 10 mu F 63 V rectangular polyester capacitors, but they are large and not exactly cheap. I considered three possible 'fixes' for the capacitor problem. Realising that I might wish to make circuit changes, I soldered terminal pins into the board at the positions marked R3, R4, R5, R6, R10, C7, C8 and C15 and then soldered the components to the pins. This avoided possible damage to the board when replacing components.

If you are using a short period vertical pendulum of ~1.4 sec period or less, substituting a 4.7 mu F polyester capacitor for C7 should be quite satisfactory. LVDTckt2

If you are using a Lehman, you may choose a much smaller polyester capacitor, down to about 0.33 mu F minimum. This acts to differentiate the displacement signal to give a velocity output. This works OK because the horizontal displacement is generally greater at low frequencies. Some additional gain may be required. A 'compromise' value up to ~2 mu F may also be tried.

However, with the terminal pins inserted as described, it is possible to rewire the board to give a long period output of up to 47 sec. LVDTckt3 The input is transferred from the -ve input of IC3 to the +ve input. C7 and C8 are removed / not wired in. A 4.7 mu F polyester capacitor is soldered between the +ve terminal pin of C7 and the +ve terminal pin of C8. R10 is replaced with a resistor of 100 K Ohm to 10 Meg Ohm to give the time constant required. The -ve terminal pin of C7 which is connected to R9, is wired to the +6 V reference line. I also soldered a reversed polarity diode across the input pins of IC3 to give 'back to back' signal input protection. This modification allows the gain of the final amplifier to be chosen independantly of the frequency characteristics by selecting the value of R9.

Modification 5

The measured oscillator frequency was 15,030 Hz with a 12.13 V supply. A 4.7 nF polyester capacitor and a 10 K Ohm resistor were supplied. Philips emphasise that the oscillator components need to have a high stability. A simple test with hot air from a hair drier showed more drift with temperature than I liked, so I changed the resistor to 20 K Ohm 1% metal film and the capacitor to 2.2 nF + 150 pF 1% polystyrene foil. Philips recommend a resistor of 18 K Ohm, but without saying why. This resistor sets the current through the oscillator switched current sources and I assume that ~this value gives the best linearity / lowest harmonic content.

Modification 6.

With a power supply of 24 V, the on board voltage regulator will dissipate 0.4 Watt and feels warm to the touch. One of the major problems with highly sensitive seismometers is that they can be disturbed by very small air motions within the usual thermally insulated enclosure and 'hot spots' need to be avoided. I fitted a U shaped Al strap between the PCB and the regulator and bent it so that the other end was in good thermal contact with the steel baseplate. This seems to have greatly reduced the temperature increase of the voltage regulator.

I have a micrometer distance calibrator which gives an accurate measurement of movement over a range of 25 mm. I used a digital voltmeter to measure the voltage between the output of the NE5521 on pin 1 and the 6 V line and moved the ferrite rod 1 mm in between each measurement. This gave me an accurate linear calibration for the output of my sensor of 307 mV / mm over a range of +/- 6 mm. The output sensitivity decreased by 10% at +/- 12 mm and it was 20% down at +/- 15 mm. Any changes / noise in these voltage levels are amplified by x213 and output. The oscilloscope trace showed a peak to peak noise output signal over 10 seconds of approximately +/-0.5 mV, which corresponds to about +/-7 nanometres. This was of the same order as the 0.305 mV per digit of my 16 bit A/D converter. Overall, I am very pleased indeed with the results.

I just wish that the environmental noise that my seismometer experiences allowed these excellent detection levels to be meaningful, but I am sure that others will be more fortunate!

Chris Chapman