I'm am attempting to use the DRV425 as a ferrous material detector and I am seeing a noise power density spectrum anomaly on ~ 20 of the 60 chips I have evaluated to date. A bi-modal distribution with about 120nT separation. See report

I pasted the report below but unfortunately none of the charts (Matlab figures) got copied (only their names).  I would be more than happy to e-mail the report (a word docx) to anyone interested.

In a nutshell, I have measured the noise power density spectrum of ~ 60 devices.  Data sets are obtained using a 1Khz sampling rate (noise density out to 500Hz), ~ 500K samples (500 seconds, noise density down to 0.002Hz), 32-channel 16bit A/D converter rated at 1/2LSB full bandwidth (250Khz) noise floor (NI_USB-6218 device) and Matlab to generate digital 69th order FIR equiripple low pass filter with 10Hz stop band and all statistical and Fourier analysis.

On the good nodes, everything matches up extremely well with the spec sheet - 1.5nv/rtHz 10 hz to 500 hz,  17nTrms 0.002Hz to 10Hz.

On the problem nodes, post filter, is see the 17nTrms noise distribution centered around 2 different "DC" fields with the two distributions about 120nT apart.  Temporally there is a "shearing" step function between the two "static" field measurements.  Since I am sampling 30 devices in parallel and only a small percentage of the nodes exhibit the shear and of those at different points in time, I am quite confident it is not an environmentally induced effect.

Any idea as to why these "problem" devices exhibit this shearing / bi-modal nature?

Note it is intermittent.  Some devices I see shearing almost anytime I look at them, some only some of the time, some (what I call the good ones) have not exhibited it throughout all my observations to date (doesn't mean they wont some day I guess).

Again, without the charts the report below is hard to read (picture is worth a thousand words) but I will be happy to e-mail the word doc with pictures to anyone interested in helping me identify (and hopefully fix) this issue.

Experiment 1 – Investigate noise source(s) in HF-system.

Set-up:

• 10 HF Nodes x 3-axis per node = 30 axis’s (A/D channels, flux gate sensors).

• A/D at +/- 5V Full Scale, Differential Input Mode with 16 Bit resolution, 1KHz sample rate.
  • 158uV LSB

• Software Low Pass Filter
  • Fpass =   1Hz (Target Transfer function within 1db of DC gain up to Fpass)
  • Fstop = 10Hz (Target above this frequency attenuated by 100db or more)

• Flux Gate Rshunt = 100 ohm
  • 4(v/v) * 12.2 (A/Tesla) * 100(ohm) * 17e-9 (Tesla rms)   = 83uV rms         ~ ½ LSB
  • Component spec.            17nTesla rms,  0.1Hz to 10 HZ      pink noise integrated over band
  • Component spec.            1.5nT / root(Hz) above 10 Hz       white noise 

• 32 Channels in DAQ, 28 attached to respective Flux Gate axis, remaining 4 attached as follows:
  • Channel 16          always shorted at screw terminals via 0 ohm jumper
  • Channel 32          always ~ 5ft long 2 conductor cable, unshielded, floating (open at far end)
  • Channel 18 and 21 grounded via 1K ohm resistor under two conditions
    • Condition 1 – 1K resistor at screw terminals of DAQ, cable back to axis disconnected from DAQ
    • Condition 2 – Cable reconnected to DAQ (~5ft long), disconnected from axis, 1K resistor across analog differential lines in connector (i.e. far end of the cable).  Cable shield grounded to DAQ A-gnd at DAQ end of cable, floating at node end of cable.

Objectives:

1. Measure Low Pass Filtered A/D noise floor                          (condition 1, channels 16 18, 21)
2. Measure Cable induced / environmental noise floor       (condition 2, channels 18, 21)
3. Measure Flux Gate Low Pass Filtered noise floor and compare against specified 17nT rms
4. Investigate nature of noise floor of any Flux gates that don’t meet spec.

Procedure:

1. For each condition, take 5 successive 50 second scans (250 seconds of total data, 250K samples / channel).
2. Filter data and analyze results as appropriate.

Data:  Condition 1

Charts below show measured standard deviation (50,000 samples / channel / chart) for each of 5 runs under cond-1.  Blue series = filtered data.

Cond-1-scan1                                                                    Cond-1-scan2

Cond-1-scan3                                                                    Cond-1-scan4

Cond-1-scan5                                                                   

Data:  Condition 2

Charts below show measured standard deviation (50,000 samples / channel / chart) for each of 5 runs under cond-2 (5ft cable on 1K ohm channels).  Blue series = filtered data.

Cond-2-scan1                                                                    Cond-2-scan2

Cond-2-scan3                                                                    Cond-2-scan4

Cond-2-scan5

Additional Analysis: 

Below shows standard deviations obtained when combining the 5 scans of each condition into a pair of data sets (250K samples, 250 seconds / condition x 2 conditons).

Problem axis’s                   6, 22, 26 and 28                 (4 total)

Intermittent axis’s           3, 11, 31                                (3 total)                    statistically significant difference between conditions

All other channels  (23 axis’s total)                           at spec ~ 0.5 LSBs             ~ 17n Tesla RMS 0.1Hz to 10Hz

Cabling add’s no signaficant noise to signal levels of A/D control channels (driven by 1K ohm)

Post filtered A/D internal noise floor (with or without external 5ft cabling) ~ 10x lower than sensors at specified operating point.

Additional Analysis:

Combine all 10 scans into a single data set (500K samples x 32 Channels) and look at distrubition of events.

Control Axis 18,                 Std = 0.06 LSBs ~ 1.8nT rms          centered about 0 LSBs

Typical Axis 8,                    Std = 0.50 LSBs ~ 17nT rms,          centered about 0 LSBs (note X bins below 10x above)

Problem Axis 22,              Std = 1.36 LSBs ~ 23nT,                   bi-modal distribution, 17nTstd at each peak but two peaks, results in overall 23nT rms

Note Bi-modal distribution, peaks ~ 3 LSBs (~102nT) apart.

Distribution around either peak is correct noise width, problem is two peaks ????

Below is the time series (over 500 seconds) of a control channel (terminated with 1K ohm, channel 18) and a Typical axis channel (channel 8),  i.e. one that is within spec all 10 scans.

And below problem channel 6 (all 500 seconds and a time zoom) illustrating bi-modal issue

Experiment 2 – Modify Gain (Rshunt) by a factor of 5x (100 ohm to 500 ohm) on typical channel 8 and problem channel 22 and examine impact.

Setup – Identical to Experiment 1 condition 2 except for gain modifications to channels 8 and 22

Objective – see if both random noise and bi-model characteristic are effected proportionately with gain change.

Procedure – Run 5 scans of 50 seconds each condition 2 with new gain on 2 channels (8 and 22)

Data:

Exp2-scan1                                                                                         Exp2-scan2

Exp2-scan3                                                                                         Exp2-scan4

Exp2-scan5

From Experiment 1

Problem axis’s                   6, 22, 26 and 28                 (4 total)

Intermittent axis’s           3, 11, 31                                (3 total)                    statistically significant difference between conditions

All other channels  (23 axis’s total)                               at spec ~ 0.5 counts ~ 17n Tesla RMS 0.1Hz to 10Hz

From Experiment 2

Add axis 4 to intermittent axis list                             (4 total now)

Channel 8 (typ) and 22 (problem) show elevated noise as compared to Exp 1 as expected.

Combine data from 5 scans into single 250K sample, 250 second data set and get stats

Channel 8 and 22 show elevated noise from experiment 1 but not at expected 5x.

• Channel 8            Exp. 1    0.50 counts                         Exp. 2    1.56 counts           3.6x  <  5.0x
• Channel 22          Exp. 1    1.35 counts                         Exp. 2    4.86 counts           3.1x  <   5.0x

Look at distribution of channels 8 and 22 over 250K samples

Conclusion – Bi-modal nature of “problem channels” associated with Flux Gate portion of chip.

250 second Time series of channel 22 (Gain modified problem channel) also illustrating the bi-modal step size increased by ~ 5x from experiment 1 (original 100 ohm Rshunt).  Again, supporting the conclusion that the bi-modal step issue is effected by the gain (i.e. within the flux gate sensor).

250 second Time series of channel 8 (Gain modified typical channel) shows no bi-modal nature but ~ 3x increase in “random” voltage noise consistent with increased sensor gain (100 to 500 ohms on Rshunt).