NE5534 v AD8031 v AD8055

[Anonymous]

To all those interested in noise,
The 5534 is definitely the best of the three for noise performance in a PI. Measured peak to peak noise at the output of the analog integrator in a Goldquest is 3.5mV, 10mV and 6mV respectively. The results are similar with a 10in search coil connected and aligned for minimum noise pickup, or with the transmitter disabled, no coil connected, and input to the amplifier circuit shorted. The readings were taken on a storage 'scope and allowed to build up over a few seconds on long persistance. It is the occasional low frequency bounce that is the culprit. You can definitely hear the difference on the headphones, with the 5534 being much smoother. I was surprised that the 8031 was worse than the 8055, as the data sheet indicates it should be the other way round. Still, I only tried one sample of each.
Eric.

 

[Anonymous]

Eric
I think that your results showing less noise from the 8055 than the 8031 are similar to my results that showed less noise from the 8031 than you would expect from the data sheet. The 8055 has a larger gain bandwidth product than the 8031 but I think parasitic capacitance is limiting the circuit bandwidth to approximately the same for either chip. And the 8055 has a lower noise figure, 6 nV vs 15.
At what pulse rate did you make these measurements?
Did you look at the output of the amplifier to see if either of these chips settled faster than the 5534? I realize that a faster settling time, which would allow a shorter delay, would only be useful under special circumstances such as, no salt water. But searching for small nuggets on dry land might be one of the special cases where a faster settling time would help.
Robert

 

[Anonymous]

In the discussion down below I wasn't only thinking about noise but also the recovery time of the IC's as Robert has just mentioned. Since on land or in a Fresh Water situation a delay of less than 10us would allow for the detection of smaller or thinner gold chains and bracelets as well as gold toe rings which are very hard to detect.
HH
Beachcomber

 

[Anonymous]

Hi Eric,
I noticed you measured the noise at the output of the integrator which reminded me of something I've been wondering about: does the configuration of the analog switches between the front end amp and the integrator significantly affect the noise ? A while ago I was looking at a design for a fluxgate magnetometer in EWWW and there's a section of the circuit which does a very similar thing to a PI detector, short pulses are fed into an integrator through an analog switch which is off most of the time. The output of the switch (which goes to the input of the integrator) is connected to ground with a 10K resistor which according to the text is to reduce noise by not having the input to the integrator floating on a high impedance when the switch is off. All the PI circuits I can think of don't do this although Vladimir Minkov posted a circuit on the geotech "tech" forum late last year which used a second set of analog switches to ground the inputs during the "off" time period.
Nick

 

[Anonymous]

Hi Robert,
I think we have to be careful with the noise figures in this application. The 8055 data certainly shows a 6nV/root Hz for frequencies above 2kHz. However, below that it rises steeply to 150nV at 10Hz. The 8031 shows 15nV above 200Hz, rising to 40nV at 10Hz. Hence, although at high frequencies the 8055 is best, at low frequencies the situation reverses. The puzzle is why, in circuit, the 8055 although too noisy for my liking, is better than the 8031. The 5534 has 4nV noise above 100Hz, rising to 10nV at 10Hz. I always reckoned that it was that it was the noise in this low frequency area that was causing the difficulties in a PI.
Your post of 31st Dec. about high frequencies being folded back into the low frequency area is something I haven

 

[Anonymous]

Hi Eric,
Search on Nyquist Folding Criterion or take a look at the url below (scroll down a way), has good pics which go along ways to understanding.
JC

 

[Anonymous]

Robert's absolutely right about HF noise folding back in the sampler. This argues for sticking an anti-alias filter between the preamp and the samplers, even if it's just a simple RC.
- Carl

 

[Anonymous]

That does make it a lot easier to see!!!
HH
Beachcomber

 

[Anonymous]

Eric
If you were testing at a pulse rate of 10 kHz then you can ignore the rise in noise at low frequencies shown in the plot on the datasheet. This is because the response of the differential integrator falls off faster at low frequencies than the noise increases. As long as the pulse rate is much greater than breakpoint on the noise graph you can ignore the 1/f noise. For the 8055 the breakpoint on the voltage noise graph is 2 kHz, so for any pulse rate much above 2 kHz you don't have to worry. This is for differential integrators only. At a pulse rate below 1 kHz that 8055 low frequency noise could be more of a problem.
Robert

 

[Anonymous]

Hi Robert,
Many thanks; that is comforting to know. Yes, I am running all tests at 10kHz. Today, I have been chasing my tail trying to do bandwidth and settling time tests. Again, without the transmitter, and feeding a sinewave signal in, the 5534 seems flat to about 100kHz, where it starts to roll off. This more or less ties up with the data sheet for a closed loop gain of 50db. Measured gain was nearer 400(52db)than the 450 expected from the gain setting resistors.
Strangely, the 8031 started to roll off at 50kHz which is odd. The 8055 was flat to 150kHz. I have gone back to the 1M feedback resistor and it will be interesting to see if there is any difference in bandwidth with the T network.
Settling time is complicated by the damping imposed by the lead foil shielding of the coil. The eddy currents induced in this will prevent much increase in speed. One would have to go to a less conductive shielding material, which will also let in more hf noise.
Eric.

 

[Anonymous]

Yes, the pictures make a reasonably complex subject alot simplier. You can see how high frequency things like radio stations etc., can get sampled and folded back into low frequency signals and really mess things up.
JC

 

[Anonymous]

Eric
There is one thing I did not show in the diagram I posted to illustrate frequency folding. Differential sampling also reverses the frequencies between DC and half the sampling frequency. In that post I used a pulse rate of 5 kHz as an example and a sampling rate of 10 kHz. So half the sampling frequency is 5 kHz. The diagram I used was for straight sampling not differential sampling. When differential sampling is used there is one final step, after all the frequencies have been folded into the range of 0 to 5 kHz then that range gets reversed. So a DC input signal gives a 5 kHz signal out of the differential sampler, and a 5 kHz input signal gives a DC signal out of the sampler. Also a 10 kHz input gives 5 kHz out and 15 kHz gives DC, etc.
So when you look at the noise plot on the data sheet you should take this frequency reversal into account. When you look at 100 Hz on the noise plot, that noise will show up at 4900 Hz at the output of the sampler. The integrator will do a very good job of eliminating that noise. What you have to worry about is the noise at 5 kHz on the noise plot. That is the noise that is going to be near 0 Hz coming out of the sampler and that is the noise that the integrator is going to let through. Also the noise at 15 kHz, 25 kHz, 35 kHz, etc is going to wind up near 0 and get through the integrator.
For a 10 kHz pulse rate the noise that will get through is at 10, 30, 50 kHz, etc. all the way up to the bandwidth of the amplifier. So the wider the bandwidth, the more noise there will be at the output of the integrator. And the higher the pulse rate, the less noise there will be.
So for a differential integrator, when you look at the amplifier noise plot, the areas that you are interested in are those near the pulse rate and all its odd harmonics out to the bandwidth of the circuit.
Robert

 

[Anonymous]

Carl
The problem with just putting in a simple filter is the spikes from the coil on-time and flyback. At the output of the amplifier those are about 10,000 times as large (order of magnitude) as the weakest signals you would like to hear. A simple RC takes about 9 time constants to decay by a factor of 10,000. So I am still working on clipping or avoiding those spikes.
Robert

 

[Anonymous]

I did not think it would be important to ground the inputs of the integrator in a PI detector. The thermal noise from a 1 meg resistor is about 100 nV/ sqrt Hz. The things that should make this negligible are that the bandwidth of the integrator is only a few Hz, so the noise is only a couple hundred nV, and the target signal at this point is already in the millivolt range, and the gain of the integrator when the inputs are floating is 1. So there is noise getting into the floating inputs but it is just too small compared to the signal and other noise to make a difference in this circuit.
However, there is one period of time when this noise might be more important in a differential integrator. That is the time when the inverting input is connected to the signal and the non inverting input is connected to a 1 meg resistor. At this time the gain of the integrator is a few hundred (in the designs that I am familiar with). I think the noise is still small but that extra factor of a few hundred might mean that it is not quite negligible.
In other circuit designs the floating inputs might be very significant.
Robert

 

[Anonymous]

Hi Roberto,
It seems that DC into the differential sampler will give DC out (actually zero). The sampling frequency doesn't even matter when differentially sampling DC. The output is zero volts.
Or am I missing something?
JC

 

[Anonymous]

Hi Robert,
Forget that last post, I am wrong you are right. Out of the differential sampler it would be 5KHz, out of the integrator it would be zero volts.

 

[Anonymous]

Robert,
Some detectors disconnect the coil to the input of the coil amplifier with an SPDT analog switch. During the transmit period and until the end of the flyback period the input to the coil amplifier is switched from the coil to ground. This removes the huge spike which is the cause of the problems you mention. the SPDT can be made with two SPST analog switches.

 

[Anonymous]

The reason given in the article for the resistor was more to do with picking up external hum and other stray noise rather than thermal noise from the resistors, I was thinking of the charge pumps typically used in the power supplies of PI detectors. I'm doing some work on my PI today so I might check it out myself and report back tomorrow !
Nick

 

[Anonymous]

Dave
Yes, that is the sort of thing I had in mind. But I have been trying to get by with just a 0 to +5 volt analog supply, and the switches I have looked at cannot handle a signal that is 0.8 V outside the supply voltage range. The single supply voltage is turning out to be a headache so I may give that up and add a -5. Then I would be able to use an analog switch.
Robert

 

[Anonymous]

Robert,
I'm not sure I agree with this. Differentially sampling a DC signal, always gives DC. In fact, always 0v.
Intuitively, differential sampling at 10kHz is just like normal sampling at 5kHz... you should be able to apply superposition. That's why a 5kHz signal ends up at DC. So will 10kHz, 15kHz, 20kHz, etc.
The advantage to differential sampling, is that low frequencies get attenuated, so a 10kHz differential sampler looks like a 5kHz single-ended sampler, with high-pass filtering. I would think that you can adjust the HP break point by varying the differential sampling delay.
I've given this all of 10 minutes of thinking, and could be totally wrong. When I get back to work, maybe I'll run some Spice sims. This is interesting to think about, because differential sampling could be used in ADC's to create a bandpass sample-and-hold.
- Carl