Timeline for PCIe, diagnosing and improving an eye diagram
Current License: CC BY-SA 3.0
18 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jul 20, 2015 at 15:48 | history | bounty ended | Funkyeah | ||
| Jul 17, 2015 at 10:14 | comment | added | Peter Smith | The WiFi chip does not need to be doing anything particular - statistically, the eye diagram you will get should not vary due to traffic, due to the fact the link is scrambled through a LFSR (sometimes known as a PRBS) which maintains constant power spectral density when averaged over a long period of time (when compared to a bit time). In this case, the majority of the traffic is probably the 'idle' ordered set. | |
| Jul 17, 2015 at 8:19 | history | edited | Peter Smith | CC BY-SA 3.0 |
added note about visualising the reflection
|
| Jul 17, 2015 at 8:11 | history | edited | Peter Smith | CC BY-SA 3.0 |
added 238 characters in body
|
| Jul 17, 2015 at 7:25 | history | edited | Peter Smith | CC BY-SA 3.0 |
Added validity note
|
| Jul 17, 2015 at 7:01 | comment | added | Peter Smith | For most testing, I just let the link run normally, and this is fine for better than 99% of all testing. Understanding the detect and polling state machines is incredibly useful for links that do not come up properly, and this is where I try and force certain states. | |
| Jul 17, 2015 at 6:58 | history | edited | Peter Smith | CC BY-SA 3.0 |
Added information about launch amplitudes
|
| Jul 16, 2015 at 19:35 | comment | added | Funkyeah | Do you force your link into polling.compliance mode to do this testing or do you just do it with the 8b/10b encoding transitions? Should the "receiving" chip be disconnected or does it need to be put in any certain mode? Right now the WiFi chip is just on and doing nothing in particular. | |
| Jul 16, 2015 at 19:27 | comment | added | Funkyeah | Few more questions if you don't mind! What do you mean when you compare 0.5V initial to Vfull and mention roun-trip time...? Are you saying that the first part is is measuring the reflection and the signal and the last part is just a signal? Is this appropriate to see or bad? Anyway to remove it? | |
| Jul 15, 2015 at 9:15 | comment | added | Peter Smith | Rolf: the caps will cause some reflection if they are a fair distance away, and it can be managed but it does need to be recognised, especially if the signal has to come from an inner layer to and from the caps, adding a couple of vias. The discontinuity is not huge, but if the link budget is being stretched, it can be the straw that broke the camel's back. | |
| Jul 15, 2015 at 8:27 | history | edited | Peter Smith | CC BY-SA 3.0 |
Corrected Gen12 to Gen2
|
| Jul 15, 2015 at 7:06 | history | edited | Peter Smith | CC BY-SA 3.0 |
Added note on capacitor geometries
|
| Jul 14, 2015 at 14:48 | vote | accept | Funkyeah | ||
| Jul 14, 2015 at 14:47 | comment | added | Funkyeah | Having said the above beautiful answer. I'm going to mark it correct even if it doesn't directly lead to solving my particular issue since it has so much useful knowledge. Any suggestions for good resources on this topic? I find the actually practical knowledge is locked away in power points and the occasional white paper. | |
| Jul 14, 2015 at 14:40 | comment | added | Funkyeah | In the comments below the post I did clarify the length is about 2.5 inches. We actually happen to be using Isola 370HR and have managed the layout quite carefully. Having said that we are also using a SOM which makes placing the caps near the transmitter essentially impossible. I am measuring on the receiver side of the coupling caps which is as close as I can get to the receiver before the routes drop into a controlled impedance inner layer. | |
| Jul 14, 2015 at 9:03 | comment | added | Rolf Ostergaard | Good answer. Maybe add a reference or some measurements where you see reflections from your caps in presence of any serious loss? I have never seen that and believe it to be more of a myth. | |
| Jul 14, 2015 at 7:41 | history | edited | Peter Smith | CC BY-SA 3.0 |
Added note about overdoing amplitudes and de-emphasis
|
| Jul 14, 2015 at 7:26 | history | answered | Peter Smith | CC BY-SA 3.0 |