How do you get rid of noise on the K20 PGA?

[argusbrown]

I need to measure some small AC signals (~100 uV) and was planning to use the PGA however the PGA seems to introduce a lot of noise in the signal path.

I have a custom board that exhibits this behavior and was also able to write a small Processor Expert program that reproduces the noise on the TWR-K20D72M demo board.

Components used:

ADC, PGA, Vref, PDB

The PDB triggers a conversion in the ADC at a 3000 Hz sample rate.

PGA Gain = 1x

VREF Mode= tight, Regulator=Enabled

ADC   Mode=differential, Conversion Time, single sample, short sample time, low-power mode disabled

V+ = V- = 0.6 V

If the PGA is enabled i get about 7 bits of noise

    1:     52                                                             

    2:    -51                                                 

    3:     54

    4:    -45                                                                

    5:     53

    6:    -45                                                                

    7:     51

    8:    -54                                                               

    9:     57                                                 

   10:    -46                                                                

   11:     55                                                                

   12:    -46

With the PGA disabled and reading DP directly I get about 5 bits of noise:

    0:      1                                                                

    1:     33                                                                

    2:     19                                                                

    3:     12                                                                

    4:     27                                                                

    5:     28                                                                

    6:     32                                                                

    7:     24                                                                

    8:     28                                                                

    9:     26                                                                

   10:     11                                                                

   11:     24

   12:     16

   13:     31

On my board I get slightly smaller values.

One thing I noticed is that when the PGA is enabled, every other sample oscillates between positive and negative. So either noise is getting coupled in from the conversion clock or I have something configured wrong (or?).

Any other thoughts? I can attach the PE project if that helps.

-andy

[argusbrown]

I adjusted the sample rate to a couple of different frequency and the noise followed the same pattern: one sample high and one sample low.

So what I ended up doing was just average the pairs together. This knocked the noise down quite a bit. I then turned on the H/W average function (to 4) and this seemed to do the trick.

With the PGA set to 64 I am now getting 28 uV RMS  (compared to almost 1.2 mV RMS of noise before).

I still feel like I am doing something wrong here.

-andy

[nicholasf]

Hi Andrew,

If you take a look at the K20 datasheet, there are some graphs and tables which may prove useful to you; I would specifically recommend looking at "Table 29. 16-bit ADC with PGA characteristics".  The ENOB lines have some great information (which I cannot copy and paste while maintaining formatting).

"Figure 15. Typical ENOB vs. ADC_CLK for 16-bit differential mode" and "Figure 16. Typical ENOB vs. ADC_CLK for 16-bit single-ended mode" are also very informative.

Here is a link to the datasheet I was looking at:

http://cache.freescale.com/files/32bit/doc/data_sheet/K20P100M72SF1.pdf?fasp=1&WT_TYPE=Data%20Sheets...

[argusbrown]

Nicholas,

Thanks for the reply. I had looked at that table and Table 27 as well. One thing that I missed is that most of the specs assume that hardware averaging is on with a minimum of 4 samples. If you turn on hardware averaging then the alternating +/1 50 bits phenomena gets averaged out. I still think there is something funky going on internal to the ADC.

So I think the solution is to simply use hardware averaging:

ADC0_SC3 = (ADC_SC3_AVGE_MASK | ADC_SC3_AVGS(0x0u));   // Average = 4

Best Regards,

andy