6035-E high frequency noise reduces resolution to 14 bits.

[GPSman]

Dear NI Tech Support and MF DAQ group,

I am not getting the full 16 bit resolution on my 6035-E and 6034-E DAQ cards due to high frequency noise.
Both the Test Panel and actual captures show 1.2 millivolts peak to peak noise with inputs grounded to AG right at the input connector in differential and single ended mode -- Test Panel screen grab and connector photo attached.
With a range of +-10V the expected +- half LSB noise specified in the manual should be only 305 microvolts (20/65536) which is four times less than what I am seeing.
So far, I have been scanning (one channel) at ten times (60 KHz) the rate I need (6 KHz) then SavGol filtering and down sampling.
The problem is that I need to scan four channels now which exceeds the the card's 200 KHz capacity and also results in huge amounts of unnecessary data.

I have been considering purchasing a new USB DAQ with better specs but am afraid that all NI products suffer from the same shortcomings and have put the purchase off.

Please share your thoughts.

Thank you.

[enderWiggin10 · Accepted solution]

Hi GPSman,

Looking at your test panel plot, it would appear that there are discrete steps in your y-axis between -0.0019 and -0.0012, or 700 microvolts.  Then you can see that there are roughly just over 2 discrete steps in your signal, which brings the value to around 300 microvolts, exactly as the resolution specifies.

You can also find that there is a noise and accuracy spec in the specifications.  You can find that for a single point, the accuracy in the +/-10V range is 1.085 millivolt.  This is on page 4.  This would allow your signal to be +/-1mV of the expected value, which the test panels also indicate.

Does this help?

best,

[GPSman]

Dear Adam,

Thank you for the reply and your thoughts on the matter.
I reviewed the specs again and confirmed your findings.
The indicated values make the 6035-E capable of only 14 bits REAL resolution (my experience), not 16 bits as the AD would suggest.
I need to see transient signal harmonics down to -120 dBm in a 3 KHz bandwidth and will have to upgrade to a 18 or 24 bit converter -- so far, oversampling, filtering then downsampling has bought me the extra resolution.

Thank you.

[ChrisR.]

Assuming you mean -120dBc, you could very well end up limited by the THD (distortion) performance of whatever board you choose, not the noise floor.  As you've seen, the noise floor can be optimized by sampling at the highest possible rate, because that spreads the noise over the widest possible bandwidth.

[GPSman]

Hi Chris,

Thank you for the reply.
Yes, dBc might be more appropriate.
My reference is at converter max of 20V PP (+17dBm across 1KOhm) (65536 counts) which would be 0dBc.
Currently, I am able to resolve components of a -80 dB 300 Hz test square wave (+/- 4 counts, 65536/8 = 8192 log = 3.913 * 20 = 78.3 dB), mixed in with converter noise at a 60 KHz conversion rate.
The seventh harmonic (0.091) (fourth line on the attached 100ms spectrogram) at -100dB is visible after SavGol filtering and down sampling by 10.
The ninth harmonic at -103 dB is just perceptible, but might as well be noise if one is not expecting it.
I'm also just able to detect a 2ms 1KHz -80 dB sinusoidal chirp and phase shifts of just one sinusoidal sample at -60 to -70 dB tops (all mixed in with converter noise then filtered down to an equivalent 6KHz rate) -- pix attached.
I have reached the limit of my hardware.

Thank you for mentioning THD which I haven't considered at all -- with very strong fundamentals, I might very well end up chasing distortion ghosts and not any real harmonic components of my signal.

Again, thank you for all your input.