12G cabling - test results and a video presentation

I presented a recent Tech Breakfast at Jigsaw24, Golden Square. Here I detail the tests we've done across four cable types and how they perform at twelve gigabits/sec (as per SMPTE 2082-1).

We've recently taken on Leader as a manufacturer of test sets and they excel in several areas - namely UHD/4K/HDR and 12G physical layer measurements. 

 all the specs for SMPTE 2082-1

I got through all the details in the video (below and on YouTube) but you can snag my results here - if you go into the 12G folder you can see the screen grabs for all the eye patterns - the filename number related to the test line in the PDF.  The Powerpoint presentation is in there too; but if you watch the video I cut all the slides full-screen as appropriate.

 The cable types are;

SD05 - Belden 1855; otherwise known as "Image 360"
SD10 - Belden 1694
SD50 - Belden 1505; otherwise known as "Image 1000"
SD73 - Belden 7731 - about the most ungainly cable you can crimp a BNC onto!

Audio over cat6a cable?

My podcast-partner in crime, the mighty Hugh Waters asked me how often I run audio signals over twisted-pair data cable in media facility builds. Has has a customer who is eager to do it this way.

I’ve done it a few times and it’s fine with a few considerations.

1. Earthing is still important and since IT people have no idea about proper grounding it can be an issue. If it’s a cat6a install in a TV facility done by me I’d have no worries, but the average IT install might have issues. But as four twisted pairs with individual screens cat6a is ideal for audio. It’s the other parts of the facility I’d worry about,
2. RJ45s don’t have the same DC/LF performance as a good old B-gauge (or Bantam) connector; if the circuits are going to be patched often I’d be wary – mechanically they aren’t great next to traditional audio connectors, 
3. From a wiring perspective; how do you nicely terminate into XLRs from a piece of cat6a?
4. Track shuffling is hard,
5. AES – just fine. Cat6a has an impedance of ~100Ω per pair, ideal for twisted-pair AES. Same observations as 1, 2 & 3.

There you go. I did work at one facility where the engineer had got obsessed with structured cabling and did everything he could over cat6 – analogue video (via baluns), audio, RS422 etc. He liked the idea that you could patch an offline i/o with two RJ45s (stereo i/o on one and video i/o and remote on the other). It didn’t work well and I put in proper cabling and patching after a year.

Sometimes convenience blinds you to fitness for purpose.

4k and UHD cabling and signal standards

I've had to dig into signal transport for 4k/UHD over the last week or so. Essentially I have a test-signal generator (SRI Visualizer TG100) running at a maximum raster of 4096x2160 at a maximum of 25 progressive frames/sec (and only 4:2:2 colour sampling; Y, Cr, Cb) with a 6G single-link output (so really 4 x 1.5G links) and HDMI 1.4 (so the same raster as the SDi). The monitors are the 24" and 30" Canon IPS 4k native monitors.

The Canon monitors will take quad-link HD/SDi and (in the case of the 24") HDMI. So, feeding the SRI single-link into a Blackmagic 4k multiplex (to produce quad-link) and then into the Canon produces four quads in the wrong colour-space!

 For an insight into what the multiplex is doing it's worth looking at the two standards for quad-link SDi. Put aside if it's 4 x 1.5G or 4 x 3G (that allows an increase to 50 or 60P OR 4:4:4 colour). But, in this case we're de-mux'ing a 6G to 4 x 1.5G signals. 

The original 4k-over-four-BNCs standard

The more recent standard; each link looks like an HD version

Clearly the converter is producing 2SI but the Canon expects SD quad-link. In fact the guys at Canon tell me they have a firmware update early in 2016 to address this. The other error is that the Canon has mistaken the 4:2:2 video as RGB - but it has at least got the raster correct.

So, what to do? Well, by throwing in another converter and taking the HDMI out of the SRI means the BM mux will get an older SD quad-link input;

This produced what we need; clearly HDMI has not concept of mutliplexed pixels and so we're now fully in SD quad-link;

tugging BNC no.4 shows the monitor is now in quad mode

The monitor gets it all right

The other thing that you have to pay attention to in "True 4k" displays (for the film snobs!) is that feeding 3840x2160 signal into a 4096x2160 monitor and letting the monitor scale-up to fill the line risks killing your resolution;

The aliasing should only the present in the top-most block, the other alias frequencies you can see here are due to my iPhone's camera!

Some very strange aliasing when a 3840-pixel line is mapped to 4096 pixels

As ever with display devices, pixel-pixel (native resolution) is always preferred

Cheap HD/SDi parts, Return Loss and the danger of short cables

Return loss is the loss of signal power resulting from the reflection caused at a mismatch with the terminating load or with a device inserted in the line. It is usually expressed as a ratio in decibels (dB); 

where RL(dB) is the return loss in dB, Pi is the incident power and Pr is the reflected power.

Return loss is related to both standing wave ratio (SWR) and reflection coefficient (Γ). Return loss is a measure of how well devices or lines are matched. A match is good if the return loss is high. A high return loss is desirable and results in a lower insertion loss. In the case of newer budget HD/SDi equipment return losses can be as bad as 12dBs (I've measure Blackmagic boxes thus) whereas the spec for 3G is 16dBs and in the case of proper broadcast manufacturers 18dBs or better is often measured (Sony, Tektronix).

I recently tested some real budget SDi parts from a reseller who is thinking of importing them from the Far East. Here are a couple of eye patterns of a 1.5G signal at the input and via the high-impedance looping output of a distribution amplifier.

If you take the first measurement as the base (it's not brilliant but that probably due to the poor signal generator) and then see what you get when the feed is connected to the DA's input and measured at the Hi-Z loop-through you see the effect of return loss - the ability of the sending equipment to drive the line impeded by reflections at the receiving piece of of equipment because of sub-optimal termination.

One thing to look out for is using short cables with budget SDi parts because the reflected portion of the signal doesn't have the chance to dissipate and interferes with the incoming signal. Replacing the short coax cable with a longer one solves the problem. This seems counter-intuitive as all engineers are painfully aware that signals get more compromised by long cables. We all carry a set of figures around in our heads; Gigabit Ethernet over cat6; 100m, 3G HD/SDi over Belden 1694; 60m, 10gig Fibre Channel over OM3; 600m etc etc. 

Matt, Wes and I were left scratching out heads with the following setup;

HD/SDi MTX, optical o/p -> 100m OS1 line -> optical->SDi converter -> SDi-HDMI converter -> TV

We were getting nothing at the TV. Checking for SDi just pre the HDMI adapter using a Tektronix WFM5200 (no eye patterns, unfortunately!) was fine, but the giveaway with sticking a signal generator just pre the HDMI converter and using the same short-length cable showed no signal at the TV. Replacing the short cable with a longer one (15cm with 1m) made it all come good. 

Clearly the Tektronix had a much better return loss on it's input and so wasn't phased by being fed with the short cable. The cheap HDMI converter was a lot more sensitive. Interestingly switching between 270MBit/s SD and 1.5GBit/s HD made no difference.

the short cable was the bogeyman!

Audio podcast 2 - the engineering!

Hugh and I continue our discussion of audio and make particular mention of cabling for TV facilities.
Find it on iTunes, vanilla RSS, YouTube or the show notes website.

Crimp tool for all current video ends?

All the wiremen want me to buy them one of these! This tool can do SD01 standard def coax (Image 360), 1694 HD Coax and the newer Din1.0/2.3 mini-HD connectors.

Video and high-speed networks - article in Broadcast Engineering Magazine

What an up-market magazine Broadcast Engineering is! Well, when they publish my stuff.
You can snag a PDF of the print version from my DropBox; BroadcastEngineering_Article_Nov2010.pdf

Cause of return loss in cat5e cable

One of our biggest suppliers asked us to test a sample of cat5e cable - we tried a couple of different RJ45 ends on the cable - a non-name one and Tyco. We ran the same 1000BaseT test on all six cores for both connector types and if you look you'll see that cores 2 & 4 consistently fail on return loss.
I initially thought it must be down to badly terminated ends but the DTX makes a distinction between return loss over the length of the cable and return loss at the remote end (how on earth it works that put is anyone's guess!) - generic RL is therefore all the reflections along the whole length of the cable that impede the transmitter's ability to send a strong signal.


Now then - it's the brown pair in every case - that suggests that the brown pair is sub spec. We didn't test to an ISO standard (because the cable isn't marked with one) so we used a generic gigabit Ethernet test which is a bit more tolerant.

I have to say I think the cable has a manufacturing fault in the brown pair.


With that in mind we stripped out some of the brown pair from core 2 (bad) and core 5 (good) and you can see the twist in the bad pair is much more variable than the twist ratio in the good pair.

So - it seems like the brown pair in cores 2 and 4 is inconsistently twisted compared to the brown in the other cores.

The great 3G cable shoot-out

Thanks to Simon Hillman for tabulating them in a way that makes trends clear. The link in the title is the directory with the A1 plots.

The tests based on a pathological signal are more extensive and test for six cable types - three HD and three SD. Since we were trying to spot trends due to cable length we feel this is the most informative of the sets of data. Simon re-did a subset using only the three HD types and 1080/50P colour bars to test for jitter which (as you'd expect) doesn't vary to any degree with length but you can see the relative damage barrels and U-Links do.

The conclusions that spring out are;

• At 3G using coax specified for HD 60m seems to be the workable cable length before attenuation becomes an issue and the eye closes below 400mV.
  
  
• SD coax goes about half the distance - this seems counter-intuitive as most SD coax has a notional analogue bandwidth (+/- 6dBs) of 360Mhz - three octave less than HD coax. Clearly the signal recovery in the WFM8300 is at play.
  
  
• The variation between the best (most expensive) and worst cable at HD before the signal becomes sub-optimal (i.e. worse than 3dBs attenuation) is less than 10m with Belden 1694 coming out on top.
  
  
• The 8300 was still able to recover a signal at 150m with Belden but only 120m with the Draka DC DVC13C. At these length the mean time between corrupt video frames would be unacceptable.
  
  
• The Condufil 1694-equivelent tracks the more expensive Belden cable very well.

As mentioned this was really a test of run-lengths for a practical guide to cabling TV facilities. In the bulk of the tests we used the correct BNC crimp connectors and the proper tools for the brands of BNCs (attached by an experienced wireman) - We did try and provoke jitter by mixing up connectors with cable but it seemed to make scant difference. It does seem for 3G HD video the newer style 4.5Ghz are to be preferred over the original HD-type cable.

In the end I suspect that these results will represent the best possible world as Tektronix gear is known to drive a coax line optimally and has excellent return loss on its inputs. Other manufactures are less so and if our experience with 1.48G 4:2:2 HD is anything to go by the massive variation in the quality of line-drivers and receivers will make these results meaningless. Nobody (particularly in these hard economic times) builds a facility with only Sony and Tektronix equipment.

Many thanks to Tom & Lee at Tek for the loan of the equipment and advice and Simon and Graham at Bryant for providing the various cable types, ends & tools. Please note all original information and test results are the property and copyright of root6

update: After IBC 2010 I had dinner with a couple of the guys from Condufil and they told me that the trick with high frequency coax is the consistency of the dialectric. The best examples use nitrogen to inflate the dialectric foam whereas budget cable is manufactured with a mechanical extrusion method. In their words this if the difference between Belden & Condufil vs Draka. This seems to be borne our by our tests.

3G cable 'shoot-out'

Three gigabit HD-SDi video has been around for a couple of years now but I've yet to see a comprehensive review of different cable types and how they handle the newer 1080p50 signals. Bryant Broadcast contacted me a couple of months ago and asked if I could conduct a test of the six cable types they recommend for HD work. So, I've borrow a brand new Tek WFM8300 (launched at IBC, not that I'd have known!) and have six drums of cable and all the BNC ends and appropriate crimp tools at the ready! I'll be measuring jitter, noise and the overall state of the eye pattern at 10m intervals from 200m down to 10m and then when we have a good feel for the various cable types we'll turn our attention to the damage that jackfields, u-links and BNC couplers do to the signal.
The image above is a screen capture of the machine with 2m of Belden 1694 cable - rated at 4.5Ghz!

Optical cabling - advice to customers

I'm not blogging much at the moment - lot on. However - this is what we're advising customers who don't use us fot their fibres;

• Grade of cable - All current models of film and video SANs make use of multimode connection. OM1 cable is still the preferred grade (62.5 micron VCSEL-optimised glass in accordance with ISO-11801) and since current configurations are 4 gigabit (moving to 8 gigabit) more attention needs to paid to circuit loss than 1 gigabit (the standard when OM1 was introduced). OM2 and OM3 cable is still unsuitable because of the 2.5dBs of loss when going between dissimilar core sizes (62.5 vs 50 microns). This is a function of current host-bus adaptors rather than the response of the cable.
  
• Bandwidth - Whereas 1 gigabit traffic will tolerate up to 8dBs of loss we are now dealing with SANs that demand at least two octaves more bandwidth and so best practise says that we now expect no more than 3dBs of loss on a SAN circuit.
  
• Style of cable - Although tight-buffered cable is easy to install it is never optimal for long runs. For interconnection between equipment within a cabinet it is appropriate and between cabinets if run in protection – Copex etc. For inter-area runs a loose-tube cable is the best solution as it is an order of magnitude more robust and although has an slightly larger install-time cost has a much lower TCO.
  
• Connectors - All contemporary host-bus adaptors and fibre-switches terminate runs in the LC connector. If existing cables are terminated in legacy SC or ST connectors they should either be re-terminated or re-run as adaptors introduce signal loss. SC or ST patch panels are fine so long as run-out cables are SC-LC (to equipment) as appropriate.
  
• Testing – We will ascertain if circuits are suitable for proposed SAN deployment by illuminating them with a calibrated laser tester (850nM wavelength, -19dB(m) signal) and measuring circuit loss – these results will be provided to the customer.

Fibre patch cords protected with Copex

A facility that we're just finishing has a lot of tight-buffered patch cords running between four bays (twelve Avids with SAN attachment and DVI extenders) - since it's impracticle to run loose-tube cable in such a restricted space Simon came up with this method of running the fibres through Copex and under the bays, breaking it out at the height of each pair of workstations - a very good solution that I'll use every time!

Fibre Channel - don't go far!

Graham asked me for some thoughts on a customer who wants to site Avid Unity SANs in several buildings and have clients connected over multi-mode -> single mode (and back) media translators. It sounded a bit dodgy so I went digging;

Clarion seem to be the most upfront about cable lengths for one and two gig fibre channel on their storage and they quote 500m and 300m respectively. In all the scenarios it isn't signal loss that's the issue (decent grade multi-mode cable degrades at about 0.5 dB /100m which makes the 500m for multi-mode eminently achievable by even modest splicing). The problem is all down to latency and since Avid expects so much housekeeping of a Unity to happen at the client (mirroring etc) you can see why all the Unity documentation quotes 150-180m (depending on which switch/HBA combo) for client-Unity connection. This (assuming C=2.5 x 10^8 m/s in gallium-doped OM1 glass) means the tolerable latency is around 600nS. This would not be achievable once single-mode media-translators are involved.

In a sense fibre channel is the worst thing to either send a long way and/or run through translation stacks (between it's native multi-mode to single mode and back again). It's Asynchronous but doesn't enjoy IP's fault tolerance and it's handshake'd but doesn't have TCP's dropped packed abilities. Worst of all it's time-dependent because it's video! It's a situation where playing fast'n'loose will end in tears.

Single mode fibre

Last week we did our first single-mode ('mono-mode') fibre job at Molinare. They are doing a job for the Beeb at Pinewood where they have to extend a dozen circuits provided by SohoNet for KVM and HD-SDi extension (the DS-Nitris editing station lives in Soho but the editor sits in Pinewood).
Anyhow - some things we discovered about single-mode over our usual multi-mode are;

• For loose tube cable you really can't tell the difference between the skinny 9-micron and the relatively fat 62.5 micron fibres because the cladding on the glass is the same width - 125 microns (see the diagram). In fact, when you prepare the ends and get them into the fusion splicer you are hard-pushed to see the difference.
• For testing you need to use a mono-mode light source and detector and the appropriate test leads. Thankfully Darren at DIS sorted me out with his tester. I'm going to look into buying one if we do more single-mode.
• The Tritec Fase-2 splicers that we use have a mono-mode config that runs the laser cooler so that the little centre-core isn't blown away by a beam intended for the bigger cores.
  

For a quick explanation of mono-mode fibre see the Wikipedia article (link in the title).

10 gig ethernet over cat5e?

It's a question a couple of people have asked me now. It is even the case that Intel and Alcatel have suggested they may have a line-conditioning chip that will allow it over sub-10m distances, but my response is;

The problem become apparent when you consider that 10gig over cat7 runs at 600Mhz (strictly speaking you need 22Ghz to carry 10gig data - Nyquist limit and all that) but 10gigE uses QAM and OFDM modulation techniques to achieve this. Now, cat5e cable is flat'ish to 100Mhz and cat6 to 250Mhz.

By the time gigE came along it was cheap enough to incorporate a QAM16 modulator and OFDM encoder on the network card and so they could start getting away with sub-Nyquist bandwidths. 10gigE takes this to another level with QAM64 modulation (similar to aDSL and DVB-T) and a verterbie decoder. But, even then it really does need the 650Mhz of bandwidth to achieve 10gig speeds over 100m of cable - the guys at Tyco reckoned that doing 10 gigE over cat5e would only ever be feasible over sub-10m distances, even with heavy-duty line-conditioning chips.
The strength of an IP stack is that it will tolerate lots of line noise and packet failures - the network card may reports that it's seeing the 10gig heartbeat but if you're dropping half your packets is it worth it?

Data Centre wiring

All of last week we were integrating a set of server cabinets for Dataupia and yesterday I took two of the wiremen to the data centre in East London where the service was to be housed. We spent the day finishing the install and during several of the hurry-up-and-wait periods I chatted a bit with one of the admins there. They host for several large corporates including a well known search engine and their machine room is enormous. I thought I'd seen big machine rooms in TV facilities but even the mighty Red Bee looks small compared to this! They have four diverse power feeds (from different providers) and four separate incoming fibre sets (again, totally diverse). I had a good look around and here are a few observations;

• All the power (16A and 32A feeds, terminating in the cabinets in C-form ends) starts in the mains room on Powerconn connectors - supposedly because those connectors lock - unusual.
  
• Despite every bit of equipment having a switch mode supply (and hence being an inductive load) every MCB in the mains room was C-rated and double the required capacity (C32s for the 16A circuits etc.) - I'm sure using correctly rated D-breakers would be better from a safety and reliability point of view.
  
• Air conditioning was via the floor - cool air forced out of the raised floor void and warm air extracted from above. Given that all kit draws air in from the front and cold air is heavier than warm air I'd have thought the TV practice of dropping cold air down the front of the bays was a better configuration.
  
• All of the techs and admins who saw Clyde and Linus at work marvelled at the numbered cables and the fact that all our cat6 was cut to exact length - I can't believe the entire internet runs of pre-made patch cords!
  
• Al the fibre I saw was tight-buffered OM1 run in Copex - what technical reason is there for that? Have they never heard of loose-tube cable?!

So all in all an interesting day - I was gratified that the way we build machine rooms in TV seems more sensible than these guys (and I'm assuming this is a tier-one provider). With all this in mind I signed up for the following at my institute. If you're interested drop me a line and come next Tuesday;

IET London, Hammersmith Section
20th November 2007 - DataCentre Design & Build
Talk by Mike Stokes of Symantec

Mike Stokes leads the data centre consulting practice for Symantec in the UK. He has over 10 years experience in guiding clients in the definition of how to meet their requirements for data centre capacity covering capacity planning, technical specification, project definition, financial analysis and business justifications.

His talk will cover the common difficulties being currently being experienced by many companies in deciding how to provide adequate space, power supply, connectivity and cooling on a 10+ year planning horizon, for IT architectures that have been changing dramatically every three to five years, a trend which is only expected to accelerate.

This subject should have something of interest for all IT Professionals, IET engineers and managers of IT operations.

Fluke DTX-1800 cable analyser

The Fluke DTX-1800 Cable Analyzer provides a bandwidth 900 MHz that supports video distribution, Class F and 10 Gigabit Ethernet. Its transmissive LCD display with backlight makes for easy viewing. Its rotary knob makes learning easy and operation simple, keeping you in the know as to what test mode is selected. The USB port facilitates high-speed transfer of data.

This is tester that we've just bought to test the big 10-Gig ethernet project we're doing - see a previous post here.

Testing for ten gigs over copper isn't yet ratified so we use a slightly ad-hoc method;
10 Gig testing should be performed to ISO11801 ClassEA Channel (not permanent link) testing using PiMF 600 patch cables. Tyco recommend using a set of 2M patch leads for 500 tests and keeping them referenced to the tested ports.
On the DTX setup it will be ISO ClassEa Ch 25N1255. This is the latest draft standard for 10G cabling system performance. There is currently no permanent link standard to work to as the permanent link requires component performance parameters which have not been defined yet.

I'll post an example XML-export in the next couple of days.

Corning Announces Breakthrough Optical Fibre Technology

Corning's breakthrough is based on a nanoStructures optical fibre design that allows the cabled fibre to be bent around very tight corners with virtually no signal loss. These improved attributes will enable telecommunications carriers to economically offer true high-speed Internet, voice and HDTV services to virtually all commercial and residential (apartment and condominium) buildings. Current optical fibre installations lose signal strength and effectiveness when bent around corners and routed through a building, making it difficult and expensive to run fibre all the way to customers' homes.

Hmm - one to watch out for. Hopefully not the big let-down that plastic-fibre was!

cat6a, cat7 and all that 10gig stuff!

I spent a day at Tyco in Stanmore doing some training on the newest types of ten gigabit network cables. Since there is no ratified standard for ethernet at this data rate (even though both Intel and Cisco have products) everyone is referring to it with different terminology. The Germans refer to the cable as cat7, the Americans as cat6a (the 'a' is augmented) and Tyco (who seem to have the biggest portfolio so far) as XG-10gig cable.
We spent the morning going over the physics of it all - the new cable is a 600Mhz channel and by QAM64 and OFDM (which I blogged about recently) signal processing techniques they can get ten gigabits per sec down one hundred metres of cable. If you use earlier cat5e or cat6 cable you are pretty much limited to sub-30m lengths.
The differences in cable and termination are sufficiently marked with respect to vanilla cat6 as to require different tools and techniques and everything is specified (even down to the sub-50N of force you can apply when pulling it into ducts). They seem to have woken up to the fact that relying on common-mode rejection as the only means of noise reduction is flawed and consequently this new cable is double-screened - the pairs are individually shielded and there is an overall screen. This is why the cable is also referred to as PIMF (pairs in metal foil).
Reasons for differences;

• Near-end cross talk is dramatically reduced by virtue of the new ends and termination tool. When properly terminated the twisted pair and shield is maintained to within a couple of millimeters of the pin on the connector. You could never achieve this with traditional punch-down methods.


• Alien cross-talk is minimised by the over-shield - cat5e and cat6 never really enjoyed this advantage.


• Inter-pair cross-talk is minimised by the foil shield around each of the pairs.
  

There is no RJ45 plug that can be crimped on - you can only buy pre-made patch cords. Panel to panel wiring is the only termination type permitted on site.
This has to be the future of data-centre wiring - we're currently doing a job that involved 600-odd circuits like this (and it's in a visual-effects company), however - I never imagined I'd have gigabit at home and it can only be a couple of years before ten-gig ethernet is ubiquitous.

Optical link hacking

I saw this article on The Register and thought it was worth commenting on. Basically they've found that you can bend a fibre and get enough signal out of the bend to reconstruct the data stream. I can only imagine that this has been tried on mono-mode cable only (mono-mode transceivers are a lot more robust when it comes to dB loss). If you've ever ended-off a fibre you know how fragile the glass is - having to open up a loose-tube cable, scrape off the coloured ident paint, attach an optical coupler and bend the fibre just enough to get leakage but not have it break would be a real feat! I know in military applications they de-tune the sender to the point where the receiver has less than 3dBs of tolerance - any mucking about with the cable makes the link fall over and an engineer investigates. That's why the MOD ban UDP traffic on their fibre networks - they know the 3-way TCP handshake requires a link that hasn't been interfered with.
I started thinking about whether you could try this with multi-mode fibre - it's a lot thicker (62.5 microns as opposed to typ. 9 microns) and so physically is a lot easier to manipulate. However, after spending last weekend measuring the response of multi-mode cables that had been only mildly abused I realised that where you can tolerate 10dBs of loss on a mono-mode fibre (even at 10 gigabits per sec) you rapidly run into trouble at even half that data-rate with multi-mode fibre.
So I think I'll consign this story to the category of you might be able to do it in the lab but the practicalities are too troublesome.