Audio analyzer channel difference

[blueskull]

I have recently acquired a Keysight audio analyzer with 2 input channels and 2 output channels.

One of its behaviors is interesting, that the 4 channel combinations (TX1-RX1, TX1-RX2, TX2-RX1, TX2-RX2) have different performances.

At the same condition (1kHz tone, 192ksps, 131072pts FFT, differential out-differential in, 50R output impedance, 50R BNC cable, 1MR input impedance, 1Vpk output, 1Vmax input range):

TX1-RX1: -111dB THD+N, -120dB THD
TX1-RX2: -110dB THD+N, -115dB THD
TX2-RX1: -110dB THD+N, -118dB THD
TX2-RX2: -108dB THD+N, -111dB THD

Background noise is -111dB, so calculated THD+N (-108dB, -110dB, -110dB, -110dB) values match measured THD+N values.

What I don't understand is the correlation between the four combinations.

TX1-RX1 is only 2db different from TX2-RX1, which means TX1 is only 2dB better than TX2.

However, TX1-RX2 is 4dB better than TX2-RX2, why is that?

Another thing to mention is input level sensitivity. RX2 is 5dB worse than RX1 at full scale, but it 1dB better at lower amplitude, peaking at -121dB, while RX1 is very stable in THD, constant at -120dB (both with TX1 generator).

Can any audio gurus explain why is this? Can it be from ADC/DAC individual error or can it be from layout?

The analog circuit and layout are both symmetric, but the clock is not. The clock buffer is closest to RX1 ADC, then RX2 ADC, then TX1 DAC, and lastly, TX2 DAC. Will that play a role?

The clock signal is 50R series terminated on die (5PB1106), but the traces don't look like 50R. The top side traces look like 100R width, but I can't be sure since I don't know the exact stack up of the PCB, nor I know the internal layer trace thickness.





I need help form the forum on why is it. If it is a design issue (like as I said, a clock issue, but that doesn't make sense as low level performance of RX2 is better than RX1), then I will have to return it before I do any reworks on this very expensive board. If this is an individual issue, then I will relinquish the warranty and solder a binned ADC in place of RX2 ADC.

Lastly, as usual, Keysight always ships some flawed products to me. This time this card came with two stripped screws and one of them protrudes from PXI bracket and scratched my M9111A.







That's certainly not what I expected as a "prime condition" used demo unit straight from KS.

[Tomorokoshi]

Failing relay contact in the attenuator chain?

Is there any documentation on the TP test points?

Probing without extender cables between the boards will be difficult.

[DaJMasta]

While I'll admit the behavior is a bit strange.... it's well within specification, they're advertising -106dB typical residual THD within an additional 3dB of noise floor for the generator and analyzer input.

Even though the blocks look identical for each channel section, maybe the internal grounding/power layout has some resonance or some additional crosstalk that only occurs between the channel 2 input and output.  If you actually compare the FFTs, especially with some averaging on, do you see any particular tones poking through in the S22 equivalent measurement, or does it seem to be more broadband noise?

[DaJMasta]

I'm saying there may be some sort of coupling through the board from the output of channel to to the input of channel two which resonates at that harmonic you're seeing.  While it's not observed on the other channel and not observed on this channel measuring the other generator, there may still be a mode where a harmonic is being conducted through other paths in the board from the channel two generator to the channel two analyzer that does not work the same way because of the physical aspects of the layout (or maybe some part doesn't have quite the PSRR of the one used on the other channel, or there's some filter element that's underperforming the other side, etc.)

Does the behavior change with a different test tone?  For example, using the second harmonic at 2kHz as your test frequency?  If it goes away then, it could be some sort of low pass response on the part of the circuit causing the issue, so you could try a lower frequency like 500Hz and see if its fourth harmonic exhibits the same symptoms by comparing the two channels.


Not sure if their measurement suite supports it, but you could also do a few hundred point sweep of the THD+N for the whole band and see if that points to a specific frequency that's an issue or if there's some other sort of response pattern for the issue.

[DaJMasta]

Impressive!  Surprised contamination could do it, but I suppose when the board area is pretty small, it becomes much easier to happen.  Maybe another can chime in on the relays, but if they're sealed, I would do it.  The Axicom IM01 relay datasheet says ultrasonic cleaning isn't recommended, but short of full immersion and agitation, I expect they'd hold up.

[Vgkid]

What adc does this use ?
I'm slowly getting into adc audio measurements , but I don't want to become an audiophile...

[tautech]

What adc does this use ?

Pics in the OP.

[Vgkid]

Thanks.