A Jig for Tether Testing


#1

Hi All:

The current tether arrangement uses Belden 1353A cable-- a single twisted-pair of 24-ga stranded wire-- to send 10Base-T ethernet bidirectionally. An ETS EBS-10BaseT-2C-ST "Monoline Balun" (ETS calls it a balun but I would call it a hybrid transformer) is used on each end to multiplex the two ethernet directions onto a single twisted pair. ETS claims a range of 100m for this setup using a single Cat-3 twisted-pair. The Belden cable is rated as Cat-5E, so it should be fine.

Previous tether testing has apparently found that tether lengths beyond 50-70m are not reliable. This may be due to changes in the electrical properties of the cable when it is immersed in seawater. Nonetheless, I'm interested in running the ROV out to the full 100m tether length, so I decided to start performing some quantitative tests on the tether system, to try to glean clues as to why it is not working over the full 100m, and how it might be improved. In an ideal world, it might also suggest how to run 100Base-T over the tether as well.

I put together a small test jig to interface the tether system (balanced 100 ohm impedance) with standard test instrumentation like a spectrum analyzer (single-ended 50 ohm impedance):


I used a couple of BNC jacks, mated to Mini-Circuits T2-1T transformers to convert from 50 ohm single-ended to 100-ohm balanced. The transformer secondaries were connected to short lengths of Cat-5e cable, which were plugged into Cat-5e couplers. 100-ohm resistors were used to terminate both ends of the cable pair not under test. Some of the hookups were done with little pins and jacks so that the configuration could be changed between testing transmission loss and measuring crosstalk.

Here's a shot of the test jig, with the two ETS baluns plugged into the couplers, and a short piece of Belden cable connecting the two baluns. The BNC jacks are connected to a spectrum analyzer- the tracking generator feeds signals to the test jig, and the jig output feeds back to the analyzer:


I ran two sets of tests on this configuration, with the baluns connected by only a short piece of cable. The purpose was to measure the performance of the baluns alone, to serve as a baseline for future tests that add a large length of tether. The first set of tests measured between 0 and 30 MHz, running out to the third harmonic of the 10Base-T baud rate. The second set of tests was between 0 and 300 MHz, to see what the performance on 100Base-T would be. I'll plot two pictures below of the 10Base-T performance.

I first setup the jig without the baluns, running a small Cat-5E patch cord between the two sides of the jig. This was used to get a snapshot of the frequency response of the test cables and the test jig, which was then subtracted off of subsequent tests to provide a normalized measurement of the balun performance. The patch cord was then removed and the baluns attached as seen in the photo above.

Here's the frequency response of the baluns:


This shows pretty good performance. The insertion loss of the baluns is listed by ETS as <0.5db -- but note that this is in addition to the loss of an ideal transformer hybrid, which is 3dB (a factor of two in power) in each balun. The marker I've set at 10 MHz shows a loss of 6.39 dB, so each balun has a 3 dB theoretical loss plus an additional 0.2 dB of insertion loss. So far so good.

Here's the crosstalk of a single balun:


The marker at 10MHz shows a crosstalk of -32.1 dB. This meets the ETS spec of >30 dB.

I'm going to go get some sleep right now, and will post my thoughts of what it all means later.

-Walt


#2

This is good work. Please keep us informed. Could you give a complete schematic of your test jig?

SherpaDoug


#3

Hi Doug:

I'll try to throw one together- you'll see it's an incredibly simple circuit.

If I were to draw it up today, the only tool I have on my computer would be QUCS. At some point I should probably download the starter version of Eagle CAD and start learning how to use that package, though I'm not sure I have the spare energy for that right now. Another possibility is to learn how to use "Fritzing", which I gather is quite popular with Arduino folks. Any recommendations?

-Walt


#4

Hi Walt,

I had never heard of QUCS. It looks quite interesting. As a consultant I get schematics in all sorts of formats. For simple things a PDF works just fine. Eagle is pretty popular and capable. A while back I looked at Fritzing but I don't remember why I decided not to pursue it. Maybe some Fritzing fans will speak up in its defense.

I mostly use an old copy of OrCAD I got in lieu of back pay from an employer that went bust.

SherpaDoug


#5

Awesome work! :D I'm definitely interested to see what the loss and cross talk is for the 50m of cable that shipped with the kit, as well as for the full 100m that you are striving to achieve


#6

Hi Doug:

Here's the schematic. I drew it out on QUCS, which really isn't made for this purpose, but when the only tool you have is a hammer....



#7

So in looking over these test results, and examining the datasheets of the various components involved, I don't think there's anything mysterious going on. It's just that a 10Base-T ethernet link just doesn't have enough link margin to handle the extra 6dB loss of the baluns and still go out to a long cable length. ETS even seems to 'fess up to this in the spec sheet for their baluns- although the features table says "Can run to 100 meters", in the specifications table they list 250 feet (about 75 meters) as the link length.

Let's take a look at the link margins for this connection, and the various components involved.

The BeagleBone uses a LAN8710A Single-Chip PHY for its Ethernet connection. Going to the specs for its 10Base-T transceiver (page 70, Table 5.5 on the datasheet), the transmitter output is 2.5V typical, 2.2V minimum. The receiver squelch threshold is 420 mV typical, 585 mV maximum. Assuming that the PHY in the laptop one is using to talk to the ROV is similar, we have a typical link margin of 15.5 dB, and a minimum link margin of 11.5 dB.

Belden 1353A Cable is rated for a loss of 6.9 dB/100m at 10 MHz. I'm assuming that this is measured driving a 100 ohm load, so the voltage-divider loss @ DC caused by the 18 ohm cable resistance (-1.43 dB) is included in this figure.

The integrated magnetics used on the BeagleBone (7499010211A) are rated for a maximum insertion loss of 1.4 dB over the range of 1-100 MHz. There is no typical figure listed, and the loss at 10 MHz should be lower than that at 100 MHz. So I'm going to pull a number out of the air, and assign 0.5 dB of insertion loss for the ethernet magnetics. This is on each end of the link, so 1.0 dB total.

The ETS baluns were measured above at 6.4 dB loss for the pair.

So the total loss across the link is 1.0 dB (magnetics) + 6.4 dB (baluns) + 6.9 dB (cable), or a total of 14.3 dB for a 100m cable. This is measured in air. Loss may be more in water, especially seawater.

As was shown at the start, the Ethernet PHY only has a guaranteed link margin of 11.5 dB, with 15.5 dB typical. So there's no way one can guarantee proper link performance, even with the cable sitting in air.

What to do about this? Well, if we want to keep the tether a single twisted-pair, and want a 100m tether length, I don't think there's anything that can be done with off-the-shelf components to make this happen. Switching from 24-ga to 22-ga wire in the tether might help marginally. But what is really needed is some gain to make up for the 6.4 dB loss of the baluns.

Here are two ways to do this, unfortunately neither of them quick and easy. Other ideas are welcome.

1.) Build a little amplifier circuit to go with the baluns, to boost the received signal strength on each end of the link. One could cut open the balun to get at the transformer inside, mount it on a small circuit board, and have a small amplifier there. The signal at this point has an impedance of 100 ohms balanced, so we would need either a differential input on the gain stage, or an input coupling transformer. Since the amplifier output is going into the ethernet magnetics of the BeagleBone, I *think* it would be OK to drive the single-ended amplifier output directly into the BeagleBone / laptop.

2.) Put the gain stage in the ROV in both directions, so a custom dongle is not needed at the laptop. This is a much cleaner solution, but it's not clear to me that the balun will handle the additional 6dB of power on the transmit side (4X in power!) without melting. Maybe only put 3dB of gain on the transmit side? I think some fun experimentation is in order.

-Walt Holm


#8

Amazing work! Very informative.

Looking at the numbers of loss you add up, arent we even very close to the guaranteed link margin with just 50meter of tether?

1.0 dB (magnetics) + 6.4 dB (baluns) + 3.45 dB (50m cable) = 10.85dB

in the rov we also have several other factors than can affect the loss outside an controlled enviroment like your tests are performed in:

-Molex connector

-human workmanship factor of assembling the cable to the balun,

-the bend-radius of the pigtail on the balun

-unrevolved conduits of the tether inside the electronics tube

food for thought....


#9

So I've been giving some thought as to how to build an amplifier to make up for the signal loss in the baluns, and if at all possible I would like this amplifier to support 100Base-T as well. So I started doing some research on 100Base-T, which then led me to taking another round of measurements of the test jig.

I mentioned earlier that I had taken a second set of measurements (not yet posted) out to 300 MHz, to examine the idea of transmitting 100Base-T through the baluns. When I looked up the specs for 100Base-T, it turns out that it does not use Manchester coding the way that 10Base-T does-- 100Base-T is far more efficient with spectrum. Here's a reference on the coding scheme used in fast ethernet- see the paragraph headed "100Base-TX":

http://en.wikipedia.org/wiki/Fast_Ethernet

So it turns out that 100 Mbit Ethernet uses 4B/5B coding to transform a 100 Mbit/sec stream to a 125 Mbaud DC-balanced stream, which is then further encoded with MLT-3 (a three-level code). The minimum period for an alternation in MLT-3 is 4 symbols, so at 125 Mbaud is would have a maximum fundamental frequency of 31.25 MHz. This is a lot lower than the 100 MHz that I had been assuming.

I did some searching for the spectrum of 100Base-TX Ethernet and found this document:

http://www.belden.com/docs/upload/zeroberrorrate.pdf

On page 5 there's a plot of the spectral signal density of Fast Ethernet. I thought it would roll off above 93 MHz, but it looks like the signal doesn't start to roll off until about 110 MHz. I'm not sure why.

Anyhow, I figured I'd redo the balun measurements over a range of 0-150 MHz, as that will give us a good snapshot of the performance over the frequency band of interest for 100Base-T.

Here's a picture of the frequency response of the balun pair. As before, the frequency response of the test jig was nulled out before taking the measurement. Remember that an ideal pair of baluns (actually transformer hybrids) will have a loss of 6 dB.


This actually isn't bad at all. Loss at the highest fundamental frequency of 31.25 MHz is still only 0.6 dB more than ideal. Even at the third harmonic (93.75 MHz) there's only about 1 dB of excess loss or so.

Here's the crosstalk plot:


Uh-oh, this is more of an issue. Remember that the baluns were spec'd for a maximum crosstalk of -30 dB. It looks like they only hold that out to about 14 MHz, which is fine for 10Base-T. At the 31 MHz highest fundamental of 100Base-T, the crosstalk is -27 dB, which is still not bad. But things start to go sour above 70 MHz or so. Crosstalk at the third harmonic (93.75 MHz) looks to be about -18 dB. Given that the loss of the baluns at that frequency is about 7 dB, the SNR at 93 MHz, just taking into account crosstalk, is only 11 dB.

So what does all this mean? I don't think it's hopeless to put together an amplifier that will boost the received power level to make up for the loss in the baluns. We'll want to design its frequency response to roll off the gain at higher frequencies where the crosstalk starts to become a serious issue. I'll show some ideas for an amplifier in my next post.

-Walt


#10

Simply amazing.

Remind me to shake your hand the next time I see you!

Eric


#11

You replied to this at 3 in the morning? Damn. You're worse than I am!

-W


#12

So I started thinking about how to build a tether amplifier that would support both 10Base-T and 100Base-T ethernet links. Here's the characteristics I came up with:

-- 100 ohm input impedance

-- 100 ohm output impedance (to squash back reflections)

-- Gain of about 8 dB or so (counteracts the baluns plus a little more)

-- Flat frequency response ( I'm not going to try to do any equalization in hardware - yet)

-- Bandwidth of about 80 MHz or so (this can be tweaked as we go along)

-- Power supply of +/- 5V or simpler

-- Parts big enough that I can assemble by hand

This list already constrains the design quite a bit. If we use a 100 ohm series termination on the output to drive into 100 ohm cable, the gain of the amplifier is cut in half. So the raw amplifier itself will need about 16 dB of gain, or a gain of 5 -- a handy number.

For a conventional voltage-feedback op-amp, a gain of 5 with a bandwidth of 80 MHz means that the op-amp will need a GBW product of at least 400 MHz, which is pretty sporty. Most of the amps in that class are in leadless QFN packages, which are ruled out because I can't assemble them by hand. In looking through the Linear Technology catalog I found one part that looked OK- the LT1818 in a SOIC-8 package.

The output voltage swing of the amplifier will need to cover 10Base-T at +/-2.5 volts. 100Base-T is less at +/-1 V. Given that our output termination cuts the output level in half, the raw amplifier output will have to swing +/- 5V for 10Base-T. No op-amp is going to do this on +/- 5V supplies, but it will get close enough. What this suggests is that running off of a single +5V might be problematic for 10Base-T. We'll start by using +/- 5V supplies, and maybe think later about testing a configuration running off of +5V only. The -5V will have to be generated by a small voltage inverter.

I'm assuming for now that the input to the amplifier, coming from the balun, is transformer-isolated from the other two ports on the balun. If that is the case we can use a single-ended input on the amplifier without worrying about needing an additional coupling transformer. I'll need to test a balun in the lab today to make sure this is so. Similarly, the amplifier will be driving into the Ethernet magnetics of the Beaglebone/ Laptop, which uses transformer coupling. So the output of the amplifier can be single-ended as well.

So here's the first go at a circuit for the amp:


I'm using an op-amp in an inverting configuration. R2 sets the input impedance since it's driving into a virtual ground. R3 sets the gain of the amplifier at 5. C6 in concert with R3 sets the bandwidth of the amp at about 80 MHz. Parasitics are obvioulsly going to affect this significantly, but we'll use this as a starting point. R1, when added to the output impedance of the op-amp, gives an output impedance of about 100 ohms. C5 is a DC block on the output. I don't think it's strictly necessary, as we're driving into a transformer, but I'm thinking of putting it in there anyways. Given the large value of C5 and R1, I don't think there are going to be any series-resonant effects with the transformer that is being driven into.

For those who are not EEs, the funny values of some of the resistors and capacitors is explained by the logarithmic values of standard components-- 1% tolerance resistors come in values defined by an E96 sequence. Do a search on "E96" in wikipedia if you're curious.

So it's not clear to me that the above circuit is going to work correctly, due to the parasitics that are involved using voltage-feedback amplifiers at these kind of bandwidths. Another option is to use a current-feedback amplifier:


Here I'm again using a Linear Technology part, that I think should support a bandwidth of somewhere around 60-80 MHz with a gain of 5 and a 200 ohm load. This uses a non-inverting topology, with resistors R2 and R3 chosen to pin the bandwidth. The values for R2 and R3 are just a guess for now, and the spec sheet for the part is not that helpful. I'm not all that used to designing with current-feedback amps, so if anybody out there is, I'd appreciate their feedback.

So now, off to buy some parts. I'm going to be building prototypes up on this prototyping board:

http://www.twinind.com/catalog_detail.php?id=425

which is absolutely genius. Apparently this is a fairly new product, and it's kind of embarrassing that nobody thought of this years ago. It uses 0603 resistors which are pretty tiny for hand assembly, but it should work out if I don't have too much coffee.

Well, that's all for now. Suggestions for improvements welcome.