Hi Expert,
mu customer use TMP400 to read CPU temp in OTN(optical transmission network), but we find that TMP400 will jump between 33 to 40C suddenly and recover in 1 mins.
could you help to review the schematic and give some advice in this application? thanks.
BR
Chi
Hi Ren,
thanks for your reply, here is the PCB for DX trace.
it is not very easy to measure current, so we test the voltage difference between DX and DP
please help to review, and if it is possible, could you also help to explain why only 33C to 40C has jump issue when noise show up.
BR
Chi
The layout is dense. I can't draw any other conclusion from this picture. Ideally, the DX traces should have GND planes on both sides to block EMI. I can't tell what is around the traces. Here is artwork from the TMP400 datasheet:
Here is similar artwork in the app note SBOA173:
On the oscilloscope, you've identified the measurement period.
There isn't obvious noise or settling issues from this view. You may only have an intermittent disruption of this signal, so you'll have to watch for that. The noise most likely correlates to another event in the system.
ren
Hi Ren,
customer use GND and power layer on both side DX trace.
also, for the picture, showing voltage between DX and DP, my understanding is DX and DP is side of PNP or NPN, so they should have linear voltage difference with temp. why we have seen this wave if it is not affected by noise? sorry I didn't get your point in your last explanation.
BR
Chi
No problem, I am happy to explain device behavior. The TMP400 is sourcing different currents and capturing the voltages associated with those different bias levels. The switching shown is the changing of currents. For a TMP400, there should be 2 currents when RC=0 (default) and 4 currents when RC=1. The switching of the currents should happen several times during the 115ms Conversion Time. When the TMP400 is not actively making a measurement, the DX lines will be High-Z, causing the observed voltage to wander. The Conversion Time and the 4 currents are described in EC Table: 
Noise must occur during the switching period shown to corrupt the temperature measurement. I previously suggested that you monitor the switching period shown for intermittent disruptions consistent with the reported temperature change.
If the temperature measurement is not being corrupted, it can also be verified that I2C transaction is not corrupted. This seems less likely given the observed measurements of 33 and 40C.
thanks,
ren
Hi Ren，
Thanks a lot for your detailed explanation. I will ask customer to monitor the switching period.
if noise occur during switching period to corrupt the temp measurement, how to avoid it? the only way is we must have GND planes on both sides of DX, my understanding is right or not? thanks
BR
Chi
Yes, the best way to prevent noise is with shielding. You can also increase filtering of Rs and Cdiff shown in Figure 11.
When increasing these values, please observe Figure 5 and Figure 6. Note that Series Resistance of the two Rs shown is summed.
It's my anecdotal opinion that it is very difficult to fix noise problems with filtering alone, but the filters are popular.
thanks,
ren
