What kind of tools would I need to analyze 1000BASE-T/gigabit ethernet waveforms?

Question

I'm an EE student and I'd like to get some visibility into the 1000BASE-T Ethernet waveform. My goal is to induce jitter to learn more about real-world error correction, possibly resulting in research down the road.

I have no experience with analyzing real differential signals, and obviously most Ethernet-specific test hardware is extremely expensive (and I have no lab access). Can PC-based scopes/logic analysers capture Ethernet and do basic decoding (8/10b)? What other recommendations do you have? This may be a long shot, but I'm aiming for a solution < $2k.

So, my question is: What kind of tools can I use to look into 1000BASE-T Ethernet waveforms that are not extremely expensive (less than US$ 2,000)?

Answer 1

To be clear, for under $2,000, you won't be able to see 1000Base-T waveforms.

Why :

• Ethernet PHY signal is transmitted over four pairs at 125MHz on each pair.

As I answered is this question Does an Oscilloscope's bandwidth need to match or exceed the microcontroller's/processor's clock speed? If you want a quick look at the waveform, you'll need an oscilloscope's bandwidth of at least 5 times, so 625Mhz.

But you mentionned jitter and its influence. So you will need at least 10 times bandwidth: 1.25 Ghz oscilloscope.

Also if you want to see the jitter influence, you'll need an oscilloscope that is able to make an eye diagram and make jitter statistics.

To conclude with the oscilloscope, you will also need a differential probe, as the signal is differential.

• Ethernet PHY signal is transmitted with a PAM5 modulation

If you need to interpret the signal, you will need an oscilloscope or a logic analyzer that is able to interpret the signal. Most of the time, this is an option (not cheap), and only sold on high-end oscilloscopes.

I've done jitter measurement on a 1000Base-KX link (yeah KX, it's a full duplex ethernet over 2 pairs @ 1GHz), I was working with a 8Ghz scope, a differential Ghz probe, etc. Also I was asked to implement an experiment with a clock source with a controlled jitter instead of an oscillator, to see the influence of the jitter.

What you can do, is to have a board were you have a full and direct access to the PHY, to be able to check the MDIO registers. Unfortunately the clause 22 (for Gigabit Ethernet) doesn't provide detailed statistics. So you need to check the vendor specifics register to see if they have implemented low level statistics like in the Clause 45 (10G Ethernet).

Edit:
A 1Ghz oscilloscope is minimum $10,000.
A differential 1 Ghz differential probe is minimum $2,000.

Answer 2

Further to Zeql's answer, I had occasion to audition a variety of communications testers in a previous life, and what you are asking is two separate things:

1 - Looking at gigabit ethernet waveforms. This is "easy", you just need a scope as described by Zeql. I'd expect a university to have something kicking round their electronics/physics/rocket surgery department that may be suitable, possibly you'd "only" have to buy a plugin module for it. Kit like this can be rented, but it's not cheap.

2 - Inducing jitter / errors: This suggests generating a test signal, which ideally is not just the job of the nearest GBE network card (of unknown stability/reliability) plus some hacked-up system to interfere with it. At these speeds, and to get any meaningful data/results, you need something made for the job.

Kit exists to do this, if you have the money, but some of it costs more than my house. It's also rentable, but not cheap.

Companies who produce this type of kit include:

• Exfo (Their Southampton, UK office are nice guys & super-helpful)
• Phoenix Datacom (their kit is terrifyingly capable, they seemed like hardcore geeks)
• Trend (they're kinda the Microsoft of the tester world)

All this said, let's step back a bit: Why do you need to investigate error-correction at a million miles an hour? You could just as easily chose 100Mbit, 10mbit or even 1Mbit from the Ethernet standards. Or you could ditch ethernet and pick some other scheme - RS485/422 is differential, the speed / signalling / coding scheme is up to you although just running a link between two PC's or micros at 9600bps presents a much slower target, well in the scope of a BusPirate or Open logic sniffer and easier to affect by tinkering.

My background being telecomms, there's a million bits of kit out there running all sorts of systems for getting bits down a length of wire, fibre, or through the air from 300 baud to Tb/s with all manner of line codings, error-corrections, etc. so you can expand the hunt as far as you like - ethernet is by no means the only application and certainly not the most interesting (or potentially lucrative).