ADRV9002 Central Frequency offset and noisy spectrum

Hi Evgeny, 

Thanks for the detailed description.

1. This may be caused by not syncing up the clk on the evaluation board to the spectrum analyser. Do you use a ref clk from the FPGA to the analyser ref input. 

2. The Tx spectrum will have odd harmonics present due to the LO. These will require filtering but are expected as the LO is a square wave and produces odd harmonics in the Tx signal. 

3. This plot again looks like there is intermodulation spurs where the modulated signal has mixed with the LO spurs, this again is an expected effect when modulating. 

4. I have not seen this on my own setup, but i will check it again and get back to you. 

5/6. Do you have a noise spec that you are working to? 

Regards

Hi Ruairí,

Thnk you for the fast reply.

I'll get a feedback from the developer.

However there are some additional questions.

1.According to the obtained results the most better sideband suppression is -70 to -68 dB. But they would like to achieve -80 to -85 dB. in the datasheet for ADRV9002 in the transmitter spec table there is a value of an OUT OF BAND NOISE FLOOR −156 dBFS/Hz. The question is what spur rejection in narrowband mode does ADI consider normal?

2. On the Auxiliary Tab there is a DAC3 option. The max value is 4095. While changing it nothing is happen. What is this and how it can be used?

Hi Ruairí,

Additional information requested on answered points:

1. The engineer is using R&S FSVR7 anf there is REF IN. However could you please confirm which output on FPGA board he should connect? Another question is could you please clarify why we shoud sync the board with analyser as the PLL is located on transceiver board? Do you think that it is just a measurement problem and the frequency was set up correctly?

3. the problem is that the central frequency was moved up on the modulation frequency value. it can't be like that in case of I/Q modulation. Why?

Regards,

Evgeny

Hi Evgeny, 

1. Correct I imagine its just a measurement issue. There is a clk signal sent to the FPGA and one back from the FPGA to the EVB so they are in sync. (DEV_CLK_OUT)
To sync this up for measurement equipment actually you will need to use an external clk source, from a signal generator and then you can sync the signal generator and the analyser and that should clean up the freq offset. 

2. This is used to trim the VCTCXO on the board. It is not supported in the TES version that you are operating on. 

3. I'll look into the modulations on this further for you. I would advise however that you move to a C0 board and use the latest TES version (currently 17.1). We have stopped supporting B0 silicon. 

Regards

Hi Ruairí

Thanks for your reply.

Could you please also provide your thoughts regarding:

1.According to the obtained results the most better sideband suppression is -70 to -68 dB. But they would like to achieve -80 to -85 dB. in the datasheet for ADRV9002 in the transmitter spec table there is a value of an OUT OF BAND NOISE FLOOR −156 dBFS/Hz. The question is what spur rejection in narrowband mode does ADI consider normal?

Regards,

Evgeny

Hi Ruairí,

in addition to my previous message I have some feedback from the engineer.

1. He confirmed that uses spectrum analyzer in a real time mode and it's not a measurement problem.

2. He enabled DAC3 function in TES 0.15 and adjusted a central frequency well. Actually the error is a minimal.

The main problem is many spurs in the spectrum. He would like to use the chip in DMR mode. According to the application the desired signal should be aplified to 60W in a linear mode. While doing this additional spurs will be also amplified which is not acceptable. So the task is to suppress these spurs as much as possible.

Regards,

Evgeny

Hi Evgeny, 

Narrow band (NB) on this part works in a low IF mode, there is some filtering on the input and we do not tune the desired signal to the zero IF because the LO noise will affect the signal. With the low IF the signal and adjacent channels are sampled and then we use a digital NCO to bring the signal down to zero IF. We then use the PFIR to filter out everything other than our desired signal. This is shown in Figure 1

Figure 1.

Figure 2 shows a simple example of this rejection happening. 

Figure 2. 

With this in mind we measured the image rejection in the narrow band mode and found that we got around 86dBc rejection. This measurement is very difficult and a wenzel blocker. 

Figure 3.

If you wanted to get some more rejection out of this setup what you could do is to increase the IF as much as possible so that band pass filter attenuates the blocker at image frequency even more.   

Hi Evgeny, 

Narrow band (NB) on this part works in a low IF mode, there is some filtering on the input and we do not tune the desired signal to the zero IF because the LO noise will affect the signal. With the low IF the signal and adjacent channels are sampled and then we use a digital NCO to bring the signal down to zero IF. We then use the PFIR to filter out everything other than our desired signal. This is shown in Figure 1

Figure 1.

Figure 2 shows a simple example of this rejection happening. 

Figure 2. 

With this in mind we measured the image rejection in the narrow band mode and found that we got around 86dBc rejection. This measurement is very difficult and a wenzel blocker. 

Figure 3.

If you wanted to get some more rejection out of this setup what you could do is to increase the IF as much as possible so that band pass filter attenuates the blocker at image frequency even more.   

Hi Ruari,

Our customer considering to move to a C0 board however they still have some doubts regarding the performance of C0 revision.

in this case they are asking you or your colleafues to reproduce their settings and whould like to see spectrum diagrams.

Is it possible to implement? what do you think?

I have attached his settings.

They need to look at the resulting spectrograms with the C0 board. It is especially interesting - do they have to work on a DAC (DAC3) on the C0 board to get into the carrier frequency specified in the program, or has this bug been fixed? Could you try to set the "Freq Offset" plus and minus - will spurs appear in the spectrum. In theory, they shouldn't

Regards,

Evgeny

DOCX