The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Design >> High-Speed I/O Design >> LVDS/ AWG28 cable model
https://designers-guide.org/forum/YaBB.pl?num=1166199032

Message started by Visjnoe on Dec 15^th, 2006, 8:10am

Title: LVDS/ AWG28 cable model
Post by Visjnoe on Dec 15^th, 2006, 8:10am

Dear all,

I have some questions about a 32MHz LVDS-like interface that drives a 1m AWG28 cable:

1. Any good pointers to a (HSPICE) AWG 28 cable model?

2. I assume that the characteristic impedance of the cable is 50Ohm. Normally, this would mean that the LVDS
TX/RX should be terminated at 50Ohm too, to prevent reflections.

I was wondering however, since the operating frequency is pretty low, if the transmission line model still holds, since wavelength > cable length. If this is true, I could terminate the TX/RX with a higher impedance (e.g. 500 Ohm) and thus save 10 times the current for the same output swing...

Any comments on this reasoning?

Kind Regards

Peter

Title: Re: LVDS/ AWG28 cable model
Post by SRF Tech on Dec 15^th, 2006, 8:48am

1 quick question before comment: What is your expected rise/fall time?
Operating frequency is not nearly as important as rise/fall time when deciding if you can ignore the T-Line/impedance matching or not.

Title: Re: LVDS/ AWG28 cable model
Post by Visjnoe on Dec 15^th, 2006, 10:45am

Dear,

the rise/fall time is about 1ns.

Kind Regards

Peter

Title: Re: LVDS/ AWG28 cable model
Post by SRF Tech on Dec 15^th, 2006, 11:30pm

To answer your quesion:

1. I personally do not know of a SPICE or equivalent T-Line model for AWG28 cable.

2. Do not assume the cable impedance is 50 ohms. Cables need to be specially constructed in a controlled manner to provide a specific impedance. Such cables include coaxial, shielded or twisted pair; but they are not just randomly twisted/wrapped, shielded or whatever, it is done by a well designed specification (or it is otherwise a cheap cable with poor specs). If you are just talking about a free form string of AWG28, it is most likley not 50 ohms and I would guarantee that the impedance will change radically over its length as the cable's positions relative to other wires/power planes/PCB/connectors etc changes. So since you are operating at low frequencies, does this matter is your next question?

Thats why I asked about the rise/fall time. The frequency components of your waveform tends to be more dependent on the dv/dt's and di/dt's as opposed to simple periodicity. A 1nS risetime will exhibit most of its frequency energy components upto and including 500Mhz (there will be some energies above that but not enough to worry about). The other issue is the length of your rising/falling transisition along your cable. By this I mean that when you have a 1nS risetime, that risetime is exhibited as a wavefront of a particular length along your cable. This length is very dependent on the propagation delay of the cable which can vary dramatically, but lets assume its roughly 5.7nS/m (it can vary greatly by medium but this is a middle of the road number for our discussion here). At 1nS, your wavefront/length of risetime is roughly 0.175m long (1nS / 5.7nS/m). Now a lot of people have rules of thumb when deciding whether they can treat a system as lumped element (i.e. ignore T-line impedances) or distributed T-line (pay attention to impedances), generally it involves this length of risetime divided into a fraction and then compared to the length over which a signal needs to travel.

In our case, your 1nS risetime, with a length of 0.175m, if we divide it by this general rule (some poeple use 4, some 6 some 10, I like 6) would state that you can ignore modeling your cable if your cable propagation delay is less than 0.175m/6 or ~30mm. Obviously your cable is far to long to ignore impedances. Given that your AWG28 is not a controlled impedance, expect lots of noise/ringing and possibly intersymbol interference and most likley a lot of EMI emissions.

I would recommend one of two options (actually I recommend both):
1. Get a better cable with a controlled impedance
2. Brain damage your rising edge.

regarding #2; at 32Mhz, your period is ~31nS...why do you need a 1nS risetime. Go with an 60/40 rule where 60% of your periodic waveform is DC level and 40% is for transitions. If such is the case, your rise fall time would be ~6.2nS. Your highest frequency content (of concern) would be ~80Mhz and most importantly your length of the rise time would be ~1.08m. Now this will still require a distributed model for your cabel, but look what you gained, your impedances mismatches will be much more forgiving, ringing and EMI emmisions will be reduced significantly, you will operate at lower power because your dv/dt's and di/dt's will be much lower and all your parasitic caps and inductances will be less of a nuisance. Your signals will still have ~9.3nS width on your eye opening, plenty for reciever resolution in even older process technologies.

Still, inorder to ignore impedances entirely for a 1m cable, your rise times in my book should be at least 6m or ~30nS, which is beyond your spec, however, by reducing the risetime as it stands, you can gain a lot and not sacrifice performance. Again though, impedance matching will be fairly important to you, still (mostly because it is a 1m cable).

There is a lot more that can be discussed, but this post is long enough; soI hope this helps you understand what issues to consisder as you approach your design.

-SRF

Title: Re: LVDS/ AWG28 cable model
Post by Visjnoe on Dec 18^th, 2006, 12:34am

SRF,

thank you for your very elaborated answer!

Kind Regards

Peter