DS280MB810: Crosstalk, Noise

Hi Nikita,

I'm investigating these 2 items internally and will get back to you.

Can you share a bit more information about the cascade design you are looking into? Can you share a system block diagram?

Best,

Lucas

Hi Lucas, 

Thank you I'll wait.

Unfortunately, I cannot share a lot about the project. But I am thinking of making a chain: TDP20x4  + DS280xxx + TDP20x4. Some mechanical parts probably might be between those ICs.

Hi Nikita,

We typically do not recommend cascading redrivers due to linearity problems, jitter accumulation, and difficulty tuning. I suggest you review this FAQ for more details and recommended alternatives.

https://e2e.ti.com/support/interface-group/interface/f/interface-forum/1368225/faq-ds320pr810-what-happens-if-redrivers-are-cascaded?tisearch=e2e-sitesearch&keymatch=redriver%20AND%20cascade#

1. Noise

Here is some additional information from a design engineer on my team.

Most of the common mode noise is as a result of TX rise-fall asymmetry and random noise. Random noise increases at higher EQ index. You will also likely see larger residue due to rise/fall asymmetry at higher gain settings. Both will result in larger AC CM noise.

2. Crosstalk

I am still investigating what crosstalk evaluation was performed on this device. I'll get back to you if I find data points.

Best,

Lucas

Hi Lucas,

Thank you for the link, very interesting, I'll reseach.

1. Ok, I see, physics is clear now, thank you! Any numbers? At least for evaluation?

2. Ok, thanks, I am waiting.

Lucas, I've read the recommendation. 

Linearity problems are mentioned then. Let me clarify one point. And let me know if better to create a separate topic for that question!

As far as I understand, re-drivers have a fairly linear section (e.g. from 0.6V to 1V, marked yellow on the picture)

Then the plot doesn't look really linear, up to saturation (also marked but very approximate, hope you get the point =) ).

So I can imagin 3 kind of Linearity problems:

1. Linearity of normal operation rates (yellow)

2. Extended Linearity that is close to saturation (oragne)

3. Case when EQ gains are tuned incorrectly and output close or into saturation (red)

Could you comment what kind of Linearity problems is mentioned in the recommendation? One, two of them, all?

Thank you!

Hi Nikita,

Linearity becomes an issue when cascading redrivers because each device applies gain to the signal. Even if the signal is within the linear region (yellow) at the first redriver, the boost applied will push the signal into the nonlinear region (orange) at the second redriver and possibly the saturation region (red) once it reaches the third redriver. This becomes an issue when transmitting training signals because it will become compressed after passing through the redriver cascade and likely won't be received properly.

If you are not transmitting training signals and don't require linearity, it is still not recommended to cascade redrivers because of jitter accumulation and tuning difficulties. In this case I recommend using limiting redrivers or retimers instead.

Can you share a bit more information about your use case? Specifically, what protocol and data rate are being used? How much insertion loss needs to be compensated at each step of the signal chain? With more information I can make better recommendations for part number and configuration.

Unfortunately we do not have characterization data on AC CM noise at higher EQ/gain or crosstalk.

We do have minimum XT characterized for DS320PR810. This is a PCIe Gen5 redriver with a similar design to DS280MB810.

Best,

Lucas

Hi Lucas, I am sorry about the delay in my reply.

>>> Can you share a bit more information about your use case?
Sorry but unfortunately, I cannot share a lot of information publicly.

>>> Specifically, what protocol and data rate are being used? 
In general, it's DP2.1 but  I've research I believe all that you have for DP2.1 (TDP20xxxx series). And I need specific functionality of DS280MB.

>>> How much insertion loss needs to be compensated at each step of the signal chain?
Formally: from -7 dB up to -22 dB. And I cannot say how it depends on frequency yet... I am under investigation. I am afraid that even if DS280's 25 dB gain is enough, it will be close to DS280's limits. And may be it's better to separate amplifying with two re-drivers. Because it looks like for CLTE Low Gain mode the linarety is better.

BTW Do you have any information regarding dependency Linearity vs EQ1/EQ2 Gain? 

Thank you for your comment about linearity! I believe I mostly found answers to my questions there.

Hi Nikita,

I understand. Would you be more comfortable sharing more information with me via E2E private message?

I believe the line rate of DP2.1 is 20 Gbps. I'm assuming the specific functionality of DS280MB810 that you are looking for is the 2x2 crosspoint. Please let me know if my assumption is incorrect.

We have TDP20MB421 which is a 4-channel linear redriver with 2:1 mux intended for use with DP2.1. However the max boost of this device is 19 dB at 10 GHz. If you find that this boost is sufficient for your application, I recommend using this part.

DS280MB810 has a max boost of 23.9 dB at 10 GHz, which should be sufficient if your max insertion loss is 22 dB at 10 GHz. Are you interested in receiving an IBIS-AMI model for DS280MB810? This can be very helpful to perform simulations with your channel s-parameters to determine if the EQ boost is sufficient.

If you are interested, please click Request Now on the product site.

CTLE stage 1 is designed to provide coarse adjustment while CTLE stage 2 is designed to provide fine adjustment of the total boost. Both stages are highly linear, however the linearity will be impacted if you cascade redrivers.

Best,

Lucas