Fluxgate Magnetometer

This was a 'research project' based closely on the renowned September 1991 Wireless World (WW) article written by Dr. Richard Noble. These notes are intended to supplement that text and may assist others attempting to reproduce his circuit.

Principle of operation

Figure 9 in the WW article shows a flux distribution pattern in which lines near to the horizontal diverge away from it as they approach the toroid from the left. My view is that the flux lines should all converge on the core, as shown in the figure here below -

The principle remains valid. A pickup coil (shown in pink) located away from the core will see the general background geomagnetic field, and it intersects eight flux lines. With a core in the saturated condition the same pickup placed over it will see the same flux. However, with the core unsaturated, as shown, the pickup intersects twelve lines. As the excitation winding switches the core in and out of saturation the difference in intersected flux (four lines) will induce a voltage in the pickup coil in the best tradition of Mr.Faraday.

If an indication only of the magnitude of the geomagnetic field were required then the task of the detector electronics would be simple. You would amplify the pickup voltage, perform amplitude detection (in the style of an AM radio) and use the resulting DC output to represent field strength. That, however, is insufficient for most magnetometers; so the circuit in the WW article is also made able to indicate the sign (direction) of the field. This is achieved by a synchronous detector. The timing diagram below may help explain how it works.

With a fairly nice square wave for the excitation drive (waveform a), any inductor would respond by drawing a triangular shape flux waveform, as shown by WW Figure 4. An inductor using the recommended R50SQ material will not draw a triangular current waveform, as you might expect from WW Figure 5. This is because of the square shape of the B-H characteristic of the nickel-iron material. When the almost vertical sides of the B-H loop are being followed only a tiny change in current is required to yield a large change in flux. Consequently, the current tends to adopt one of two discrete values depending on the drive polarity. This is shown here on the right. The top trace (1) is the drive, and the bottom trace (2) is a current probe measurement which scales to 100 mA per division. Only at the saturation points does the current increase above a few milliamps.

Points p) and q) show where saturation begins without an external field. When an external field is applied then saturation begins at point l) which is earlier than p); that is to say it is 'advanced'. However, saturation begins later than point q) at point r); that is to say it is 'retarded' or delayed. The 4016 analogue switch ICs sample (synchronously detect) the pickup signal at points n) and o) etc. This, when amplified and filtered, gives the signed DC output you require.

Circuit modifications

As far as I can see, none of the editions of WW later in 1991 contain errata relating to the magnetometer.

• I had to increase R2 to 6k8 in order to bring the frequency down to 720 Hz. Your mileage may vary. C1 (100 nF) was not drawn on the WW diagram.
   
• TR1 and 2 were specified as 2N557; but these TO5 devices are not, and never were, common. Obtaining them today is difficult. BC557s, on the other hand, are popular, easy to obtain and work well enough. Likewise BC547s. Any general purpose devices should perform OK in the driver.
   
• I left C15 in circuit (connected at the sensor head) but, as far as I can see, it does nothing. Moving the R50SQ core around its saturation states is so high loss that C15 and the core will have a Q-factor close to unity.
   
• The width of the pulse on pin 11 of the 4011 was 25 μs. This seemed rather short in relation to the pickup signal, so I increased R4 to 2k7. As expected, this increased the output of the circuit. I'm not sure that the signal to noise was improved, however.
   
• The excitation drive circuit will inject switching noise into the supply rails, if given a chance. Reasonably large electrolytics (47 μF) help soak it up. In addition, all the chips were given 100 nF decouplers.
   
• The Helmholtz calibration coils were constructed as per the text. According to the Wikipedia article the calibration factor for it should be 391.7 μT A^-1. So, for a reversal of 0.1277 amps the field will change by 100 μT, for which we can adjust VR1 to give a swing of 5 volts (±2.5 V). In the UK, the geomagnetic field is about 47 μT.
  
  The text calls for the wire ends of each section of the coils to be passed through holes 9 mm apart. My maths says that this is room enough for 45 turns of 0.2 mm diameter wire. Only 24 turns are called for. Strange.
   
• If you intend using the circuit as an electronic compass then no further amplification is wanted past the test point. Disconnect R9 and pin 3 of the 4016 from pin 2 of the op-amp and connect a 2M2 resistor from pin 8 of the op-amp. You then use the second stage just for additional low pass filtering.
  

Construction

The photo here to the right shows the almost complete magnetometer circuit. It is not essential to use a full ground plane construction method (at 720 Hz that should be overkill). I wanted to take no chances with the noise performance.

The interconnections are single core wire wrap kynar. I'm a fan of this for high density boards. Power lines require heavier gauge flex.

Since his death, obtaining the type 7a cores from Dr.Noble is now impossible. I have no good solutions to this headache. Magnetic Metals also manufacture 'square 50' tape wound cores, but distributors are no easier to find. Industry practice is to manufacture cores to special order - impractical when you just want one.

The pickup coils were wound over a sleeve made from an acetal block. I actually used 1.25 inch thick stock, which leaves a little material under all of the core, making the part somewhat stronger.

This shows the pickup coils being wound over the sleeve and core. Note the special plate fixed to the sleeve to locate it on the winder tailstock. At the headstock end a wooden block couples the sleeve to the winder drive. Winding rectangular form coils is never easy, particularly one where the width is far greater than the height. Unfortunately (as it turned out) the traverse mechanism was badly in need of cleaning and oiling, so the turns piled up unevenly. That apart, this operation went well.

Where the coil leaves the winding channel within the acetal block and crosses the gap above the core unsupported it then tends to splay outwards a bit. If you are fastidious about the winding then a couple of 'D' shaped plates could be mounted in this area to maintain continuity of the channel.

Testing

The trace on the right shows the pickup coil 1 waveform. Noble observed noise or jitter on parts of this, but that wasn't very obvious when I looked at it. Triggering phenomena?
Looking closer in at this reveals the trace shown here. If the field is reversed then what happens is that, rather than leading with a positive going pulse, you get a negative going pulse followed by a positive going one. The mid-screen spike is feedthrough from the sampling pulse.

The signal to noise ratio I obtained was not as good as the WW article suggests it should be. Perhaps an electrostatic screen around the sensor head would help. Perhaps the lead to the head should have been longer than 1 metre to prevent noise from the electronics feeding back into the head. It's curious that Noble didn't mention that point.

The magnetometer is still good enough to use as a compass, though. Channel 1 in the screenshot shows the Helmholz coil current (scaled to 10 μA per division. That should produce a swing of about 7.8 nT peak-peak. It's roughly what Noble claims is the noise floor. Channel 2 is the magnetometer output averaged over 128 sweeps - an hour's worth of data. With some imagination you can see that the field has been resolved.

In the horizontal plane I manually rotated the sensor head by increments of 45 degrees at 10 second intervals to get the figure here on the right. This compass rose was helpful here. You see that the channels are 90 degrees apart.

So, putting the oscilloscope into X-Y mode, using channel 1 to provide the X signal and channel 2 to provide the Y-signal, I rotate the sensor head by 360 degrees to obtain the trace here on the right. The 0.3 Hz low pass filter restricts rotation to a few degrees per second if reading anomalies are to be avoided.

One other problem that surfaced during testing was that there seemed to be some non-linearity with the analogue switch pins 4-3. In the end, I ditched it anyway, so this wasn't pursued further.

Conclusions

It's encouraging that such a simple circuit proves effective in measuring fields many times weaker than the geomagnetic. Still, I feel that there is further room for improvement. A beefier (and more power hungry) excitation drive would be the first shot. That (as Noble hints) might give better defined transistions into saturation.

Secondly, the very short sampling period leaves me uneasy. Some research is required to find out which parts of the pickup waveform bear useful information.

Thirdly, all the analogue signal processing is rather passé in the third millenium. DSP design anyone?

[↑ Richard's Home Page] [↑ Projects Page]

Feedback to: richard at clarcana.info
Last modified: 2007 June 23rd.