DS90LV004: DS90LV004 LVDS input failsafe biasing to '0'

The ±100 mV limits are copied from the LVDS specification; the actual switching threshold is likely to be much nearer to 0 V. But the datasheet does not actually have better information.

I find this information page 4 :

Yes, the typical value is 0 V. But we don't know how much it moves due to manufacturing variances and temperature; there is no guarantee.

Ok, so the minimum differential voltage to be applied to ensure correct polarization regardless of manufacture is +/- 100mV ? 

Because I use 16x LVDS lines spread over 4 drivers. By initially setting a 10k pull-up resistor to 3V3 on the positive input and a 10k pulldown resistor to GND on the negative input, I had 16x logic levels at '1'.

But in the end, we need to bias the inputs to logic '0'. But in the end, we need to bias the inputs to logic '0'. By inverting the resistors (10k pull-up on the negative and 10k pull-down on the positive), I have only 7 out of 16 channels in the low state, distributed inconsistently.

I made the following measurements: 

I've just tested this by changing the value of the pull-up/pull-down resistors to 1k47 from the previous 10k, and it seems to bias the LVDS outputs well in the low state.

However, this resistance value does not match the datasheet recommendations (between 5k and 15k).

What can we conclude from this? Did I miss some point?

You did not miss anything.

In order to get ±100 mV over the 100 Ω termination, you need ±1 mA. The datasheet does not give you enough information to allow you to use larger resistors.

Hum ok, I don't quite understand your answer. How can we correctly polarize +/-100mV with the range of values recommended by the datasheet between 5k and 10k ?

With the information provided by the datasheet, you cannot.

Either use resistors for 1 mA (lower than 5 kΩ). Or use 5 kΩ and hope that the actual limits are smaller than ±100 mV.

Okay, so there's an incoherence in the datasheet? How can I assure my customer that changing the value to 1k47 from the previous 10k correctly corrects the low-state polarization?

Is it possible to provide feedback for datasheet correction if the 5k to 15k information is not correct?

Hello,
Do you have any comments/suggestions for mounting a pull resistor outside the datasheet recommendations, in order to close this topic?

The datasheet refers to the AN-1194 Failsafe Biasing of LVDS Interfaces application report. It says that National Semiconductor LVDS receivers (like the DS90LV004) typically switch at −30 mV, and recommends resistors for a bias of +25 mV. The note at the top of p. 4 says that 100 mV bias might be appropriate for very noisy environments. So there is no general recommendation; this depends on the application.