6259 AO & AI phase jitter

[Oldi]

I need very accurate in phase signal generation and AD sampling. I'm using PCI6259, DA with DMA writing samples (10kHz sine wave) at 1MHz and AD sampling at 1,25MHz with onboard FIFO. AD output is connected to AI, and phase is measured with Amplitude and Phase Spectrum.vi with 10ms long samples package for Phase Spectrum analyze. Results have random error of approx. +-0.0007 degree, but after averaging, error is approximately +-0.0002. Problem is that phase have slow jitter, phase slowly shift for approximately 0,002 degree. Period of shifting depends from something on hardware. I have 3 PC, old ones P4 2,8GHz and 2 the same 1 years old 4 core Intel with 3 pieces of PCI6259. Difference between them is in period of phase shifting. On 1 quad core period is around 20 minutes, on another 2 computer period is much faster, approximately 1 minute.

Generated DA signal is complex with few different amplitude and frequency, so I must use DMA. During described phase shift there is no changes on DA samples.

Probable the reason is in nature how DMA samples are writing in to DA chip? Is there any solution to eliminate this shift?

[Oldi]

Long time analyze of that phase shift show, that there is no rules, sometimes there are 1 or 2 swings in few minute, then almost nothing in 10 minutes. Same result on all 3 computers.

Shift for 0,002 degree at 10kHz mean 0.5ns, so probable there is some sort of random propagation “noise” in a path from PCI6259 onboard clock to AD and DA converter. So probably I reached hardware limits.

[Oldi]

During the night analyze show that phase shift change with room temperature.

Legend

White and red trace: phase shift of 2 adjacent FFT 10ms package averaged in 0.2s period.

Green trace: difference between white and red

Time scale is in 0,2s per point, so 320.000 mean 17.78 hour.

Peaks is because of other tasks in Windows.

[Zody]

Hi Oldi,

Could you please post the VI used for testing? I have the same device and I will run some test to validate if it has the same behavior.

Best regards,

Ion R.

[Henrik_Volkers]

---

@Oldi wrote:

Long time analyze of that phase shift show, that there is no rules, sometimes there are 1 or 2 swings in few minute, then almost nothing in 10 minutes. Same result on all 3 computers.

Shift for 0,002 degree at 10kHz mean 0.5ns, so probable there is some sort of random propagation “noise” in a path from PCI6259 onboard clock to AD and DA converter. So probably I reached hardware limits.

---

1/10k * 1/360 * 0.0002 ~5.5 e-11 = 55ps    

Record the temperature and supply voltage of the timing circuit ... you migth find correlations 😉

[Henrik_Volkers]

BTW

How about the residual noise ?   Tested your phase measurement against analytical signals (math sine + random noise + slow bias drift) ? 

Have you tried the Extrect single tone detection.vi ?   

Or have you tried a linear least square sine fit?

[Oldi]

I'm working with Amplitude and Phase Spectrum.vi which is (for me) best method to measure phase, isn’t it.

I found solution, measuring phase of generated signal at the same time, of course at the half sampling speed so I have more random phase shift (noise), but much better stability.