Oscillation in sample data when near full scale

Are you using a default or a custom profile ?

Do you see this behaviour only at this particular frequency? 

How is the frequency domain response when you see this issue?

Is the AGC turned on ?

Hello Vanitha

1) We are using a custom profile generated by the AD9371 Filter Wizard tool. I have tried with two separate profiles. The behaviour is the same on both. Note that we changed the VCXO on the evaluation board to a 125MHz VCXO to meet our customer requirements. Here is an example of one of the two profiles that I am testing with:

<profile AD9371 version=0 name=Rx 100, IQrate 125.000>
 <clocks>
  <deviceClock_kHz=125000>
  <clkPllVcoFreq_kHz=10000000>
  <clkPllVcoDiv=2>
  <clkPllHsDiv=4>
 </clocks>

 <rx>
  <adcDiv=1>
  <rxFirDecimation=2>
  <rxDec5Decimation=5>
  <enHighRejDec5=1>
  <rhb1Decimation=1>
  <iqRate_kHz=125000>
  <rfBandwidth_Hz=100000000>
  <rxBbf3dBCorner_kHz=100000>

  <filter FIR gain=-6 num=48>
  -4
  -17
  23
  41
  -53
  -95
  117
  184
  -221
  -331
  389
  553
  -644
  -888
  1026
  1387
  -1608
  -2158
  2580
  3485
  -4786
  -7622
  9629
  32372
  32372
  9629
  -7622
  -4786
  3485
  2580
  -2158
  -1608
  1387
  1026
  -888
  -644
  553
  389
  -331
  -221
  184
  117
  -95
  -53
  41
  23
  -17
  -4
  </filter>

  <adc-profile num=16>
  527
  378
  201
  98
  1280
  477
  1587
  269
  1330
  102
  807
  28
  48
  40
  23
  187
  </adc-profile>
 </rx>

 <obs>
  <adcDiv=1>
  <rxFirDecimation=1>
  <rxDec5Decimation=5>
  <enHighRejDec5=1>
  <rhb1Decimation=1>
  <iqRate_kHz=250000>
  <rfBandwidth_Hz=200000000>
  <rxBbf3dBCorner_kHz=100000>

  <filter FIR gain=6 num=24>
  -119
  284
  -28
  34
  149
  -287
  390
  -296
  -517
  1496
  -3974
  22853
  -3974
  1496
  -517
  -296
  390
  -287
  149
  34
  -28
  284
  -119
  0
  </filter>

  <adc-profile num=16>
  446
  354
  201
  98
  1280
  711
  1679
  818
  1500
  716
  853
  21
  43
  37
  25
  200
  </adc-profile>

  <lpbk-adc-profile num=16>
  561
  362
  201
  98
  1280
  281
  1539
  143
  1343
  57
  822
  35
  48
  40
  24
  188
  </lpbk-adc-profile>
 </obs>

 <sniffer>
  <adcDiv=1>
  <rxFirDecimation=4>
  <rxDec5Decimation=5>
  <enHighRejDec5=1>
  <rhb1Decimation=2>
  <iqRate_kHz=31250>
  <rfBandwidth_Hz=20000000>
  <rxBbf3dBCorner_kHz=20000>

  <filter FIR gain=-6 num=72>
  0
  2
  3
  2
  -3
  -12
  -19
  -11
  15
  50
  68
  37
  -48
  -149
  -189
  -98
  122
  362
  444
  226
  -264
  -776
  -938
  -481
  516
  1551
  1894
  1017
  -968
  -3154
  -4127
  -2643
  1642
  7791
  13834
  17574
  17574
  13834
  7791
  1642
  -2643
  -4127
  -3154
  -968
  1017
  1894
  1551
  516
  -481
  -938
  -776
  -264
  226
  444
  362
  122
  -98
  -189
  -149
  -48
  37
  68
  50
  15
  -11
  -19
  -12
  -3
  2
  3
  2
  0
  </filter>

  <adc-profile num=16>
  589
  351
  201
  98
  1280
  112
  1505
  53
  1354
  21
  834
  40
  48
  41
  24
  190
  </adc-profile>
 </sniffer>

 <tx>
  <dacDiv=2.5>
  <txFirInterpolation=1>
  <thb1Interpolation=2>
  <thb2Interpolation=1>
  <txInputHbInterpolation=1>
  <iqRate_kHz=250000>
  <primarySigBandwidth_Hz=75000000>
  <rfBandwidth_Hz=200000000>
  <txDac3dBCorner_kHz=187000>
  <txBbf3dBCorner_kHz=100000>

  <filter FIR gain=6 num=16>
  -44
  -227
  232
  -225
  -5
  925
  -3253
  21237
  -3253
  925
  -5
  -225
  232
  -227
  -44
  0
  </filter>
 </tx>
</profile>

2) I tested now also at 2500MHz LO. The exact same oscillation happens at 2501MHz. If I set the signal generator at 2510MHz, I still see the same oscillation (as seen below in the screen grab from IIO Oscilloscope):

3) The noise floor of the frequency domain response starts 'bouncing' each time an oscillation happens, indicating the discontinuities and high frequency content of the oscillation. The following is the frequency response before oscillation starts happening:

The following shows the noise floor after oscillations start happening:

I noticed today that at the point where the oscillation starts happing, the RSSI level reported toggles between 0 and 1.25dB. Perhaps this is a clue as to what is happening inside the AD9375.

4) This is in manual gain control mode, not AGC. Here are some screen shots showing the AD9375 configuration inside IIO Oscilloscope:

I hope that helps with diagnosing what I going on. Like I said, the only change we made to the AD9375 evaluation board was to replace the VCXO with a 125MHz functionally equivalent one.

Thanks

Gavin

Can you please probe the DevClk pins manually on the AD9375 board and make sure that the DEV_CLK frequency is correct and within specs as per datasheet.

Here is a screen shot from the oscilloscope.

The device clock is 125MHz. The mean peak to peak voltage is 0.46V. These are within the requirements.

What is the gain index used in the MGC case?

Can you please check if this issue is observed in AGC as well ?

The AD9375 RX gain is the default (30). I will test if it happens in AGC as well.

It does also happen in AGC mode. Obviously it only happens at a much higher input signal level because the AGC algorithm increases the attenuation of the AD9375 RX input to maintain a constant 6dB RSSI level. So it only started happening once the AGC algorithm had run out of range on the RX attenuation input (maximum attenuation 30dB). This corresponded to about a +20dBm input level. 

Do you have LNA in your setup? What is the gain index used in the setup? 

The plot shared looks like the Rx is getting saturated. The recommended input power would be -14dBm as per the datasheet and we typically recommend customers to operate at this power level .

Hello Vanitha

I am testing on the ADRV9375 evaluation board which simply has a balun and impedance matching.

I agree, this does happen when the Rx is saturated. I measured now on the evaluation board and the saturation / oscillation starts to happen at around -6dBm. So this is exceeding the maximum recommended CW signal level of -14dBm.

So does the -14dBm input level mean at the input to the ADRV9374 evaluation board?

I find this interesting. At -14dBm, the time sample data from the AD9375 is only +/-11000 roughly. This seems to be a waste of the available AD9375 dynamic range considering that the digital range is +/- 32767.

Do you know why saturation in the AD9375 results in oscillation?

Thank you.

Gavin

gavint said:

So does the -14dBm input level mean at the input to the ADRV9374 evaluation board?

Yes at the RX input of the board.

Will get back to you on the other queries.

gavint said:

At -14dBm, the time sample data from the AD9375 is only +/-11000 roughly.

AD9371 is a sigma delta ADC  . At -14dBm input  level how much is the power level in dBFS?    

How do you say the sample data is +/-11000? What are the number of samples in the RX while capturing data?

When the power level is 0dBFS you can change the number of samples to 32768 in the RX data plot tab and you should be seeing data of +/- 32768 samples in the RX capture . 

gavint said:

At -14dBm, the time sample data from the AD9375 is only +/-11000 roughly.

AD9371 is a sigma delta ADC  . At -14dBm input  level how much is the power level in dBFS?    

How do you say the sample data is +/-11000? What are the number of samples in the RX while capturing data?

When the power level is 0dBFS you can change the number of samples to 32768 in the RX data plot tab and you should be seeing data of +/- 32768 samples in the RX capture .