Page 6 of 6
I read with interest your articles on network streaming and USB to S/PDIF converters and about the weird variations in jitter you saw on your analyser vis. dirt and reversing the connections on the optical lead. I think I can throw some light on these, excuse the pun.
The TOSLINK is basically just a fibre optic cable – but in the fibre optic timeline it’s pre-stone age. It consists of a glass or plastic core of approximately 1mm diameter, covered by a cladding of more glass or plastic of a higher refractive index. This protects the core from scratches and contains the light in the core. The refractive indices of the core and cladding are chosen such that total internal reflection occurs over a wide range of angles that the light rays make at the core/cladding boundary. This ensures that the maximum amount of light makes it to the other end. The light source is a red LED, which means that a spread of wavelengths from about 650nm to 850nm are introduced into the core. At the time of TOSLINK’s introduction, a LED was the only practical source, its’ large emitting area dictated a large core fibre unless costly bulk optics are used.
The problem is, the larger the core of an fibre optic cable, the worse its’ performance! Rather like a waveguide, large cores allow large modal spread – from the ray that theoretically travels down the axis of the fibre to the one that bounces off the core cladding interface thousands of times. The response to a step function – the LED turning from off to on for example – will cause light at thousands of different modes and several different wavelengths to be sent down the fibre, and of course they will arrive at the other end at a range of different times. The effect at the detector will be to turn the step into a slope.
This slope (neglecting electrical effects in the detector) is thus caused by modal spread and varies according to how many bends and coils are in the cable. If you can measure optical power with your analyser you will have noticed how sensitive this is to bends in the cable. This modal effect limits the theoretical bandwidth of TOSLINK cable to about 10Mbps per metre. If four or five turns of cable are wrapped around a, say, 100mm mandrel (piece of plastic pipe?) the high order modes will be dispersed into the cladding and only the low orders will make it to the end. The optical power will be reduced considerably but the recovered signal should be a lot cleaner and – in theory at least – less jittery.
Sorry for the lengthy preamble but the effect that you’re seeing is I believe caused by a phenomenon known as return loss. TOSLINK connector ends are flat polished, which means that some of the light gets trapped inside the cable and bounces around with a whole new set of modes. Some of these modes escape at both ends to be reflected by the LED source and the sensor back into the cable again. This effect is strongly dependent on the cleanliness and condition of the connector end faces – the two ends are never going to be exactly the same. Cleaning, polishing and swapping the cable ends over will affect the magnitude of this return loss factor. This adds to the modal chaos in this large fibre and I believe will be measurable as jitter.
Large cores are good for tolerance to dirt and misalignment in connectors, but bad for everything else. The industry standard multimode fibre in general use has a core diameter of 50μm – 20 times smaller than TOSLINK. It’s bandwidth is typically 1Gbps per kilometre (a bit different!) So-called single mode fibres – the kind that our information superhighway is built on and the Government and BT baulk at the cost of installing – has a core diameter of 9um and a bandwidth 50 to 100 times as great as multimode. You get huge bandwidth and huge distance, but the quality of the installation and components has to be high. So-called ‘fibre to the home’ has floundered on the component cost required to do this reliably. The light sources are lasers though, which have the advantage of being both monochromatic (usually IR, 850nm to 1500nm) and of very small cross sectional area and can thus inject into very small cores. Considerable efforts are made to minimise return loss as well, by using angled, domed ended connectors. A short multimode cable isn’t necessarily better than a long one, as the long cable increases the chance of high order modes being dissipated. It may be therefore, all other things being equal, that a 10m cable sounds better than a 1m one.
Apart from DVI/HDMI, I can’t think of another digital protocol that is time contiguous – data is always time stamped and packeted these days. And as we don’t see in the same way as video is transmitted, I think this makes audio unique – and uniquely vulnerable!
It would be great to make an optical link using single mode fibre, laser source and narrow band detector. The parts aren’t hugely expensive, but too expensive for manufacturers, I expect (pounds, not pence!). And TOSLINK isn’t going away because the automotive sector have picked up on it – because it’s cheap!
The telecoms sector fibre optic market eclipses all else into insignificance. As a result it is surprisingly difficult to optimise fibre optic systems that don’t fit into its’ value set. In general manufacturers are not too interested in making low volume ‘specialty’ parts at marginal profit when they can make a far bigger one in the telecoms business. I suspect this may be a reason why TOSLINK has had such a long life – it’s not economic to replace and in bare data rate terms, has sufficient capacity.
Hope this makes some sense at least. Please continue with your investigations as I think you’re breaking new ground here.
Audioquest Cinnamon optical digital TOSLINK cable uses a high purity polymer fibre with low dispersion to lessen jitter, they say.
Thanks for that detailed insight Keith. It saves us all from a lot of unknowing speculation on a topic that is obviously well known about outside audio – and its Stone Age cables and connectors!
The irony here is that digital cables do measurably affect audio quality, when it has been firmly believed until now that they cannot do so by the very nature of digital. Since the restricted bandwidth of long electrical cables introduces jitter by lengthening the zero crossing transition it seems we are dammed in both camps! The following letter on the problems of Cat5 adds to our woes in this field, but also points again to fibre optics.
I can’t help suspect, however, that because the audio market will pay for a good product, better fibre optic cables and terminators will become popular in due course. I know they are appearing now and I expect somewhat more mail about all this very soon! NK
I was interested to read Noel Keywood’s article ‘Media Message’ in the May 2012 issue. I was particularly interested in his comments regarding signal degradation due to passage over network cables. As a computer network engineer, I am well aware of signal problems due to the poor shielding qualities of standard (CAT5) network cables. Poor cable installation is a very common cause of poor network performance, and can be hard to diagnose unless you know what to look for and/or have the right test equipment. So if you are using network cabling in your hi-fi system, the cable installation needs to be considered carefully.
CAT5’s ability to prevent external noise interfering with the signal, comes from the signal leads being twisted in pairs; each signal lead twisting through 360 degrees continually along it length. This means that if wireless noise hits the cable in one orientation, it will also hit the cable in a 180 degree opposite orientation very nearby. Thereby, any alteration of the signal caused by noise in one orientation, is cancelled out by the same noise passing through the cable in the opposite orientation nearby.
Relying on twisted pair signal leads, to prevent interference, is not a good way of avoiding signal degradation, but for the vast majority of network installations it is good enough. The leads are cheap and easy to install, which leads to good enough being just that - good enough.
So what do good network cable installers do to avoid signal degradation? The key is maintaining the twists. As the twisting provides the only shielding, if the twists stop, so does the shielding.
So what are the common causes of untwisting:
Over-tight cable ties. Cable ties can pinch the cable, causing the twists to straighten out. A good rule of thumb is that you should be able to push the cable back and forth through the tie. If the tie pinches the cable so tight that you cannot move the cable through the tie, then it is also tight enough to straighten out the twists, and you’ve lost the limited shielding twists provide.
Over-tight bends. Cables need to go round corners in smooth curves. If you see a cable pulled taught around the corner of a wall, or wrapped into tight bunches, then there is a good chance the twists have straightened in the bend.
Too much bare wire at joints. This is common when people make up their own cables. The cables need to be untwisted to enter the joints (end connector, punch boards, and mounting plates). Where the cable is straightened shielding is lost, so it needs to be minimised. For example, the straightened cables need to be totally within the end connectors. If you can see bare signal wire entering the connector, then the connector has not been made up properly and signal degradation will result.
Another thing to consider is avoiding noise sources. Try to avoid laying CAT5 network cables alongside or past RF noise emitters:
Do not run network cables through the same conduits as mains cables. Mains cable emit electrical noise. Running a main cable alongside a CAT5 network cable is asking for trouble. If the network cable has to cross a mains cable, try to do it at 90 degrees.
Where cable runs pass though ceiling spaces, keep them away from fluorescent lighting units. Avoid passing cables near to electrically noisy devises such as refrigerators.
However, there is one simple way of overcoming all these problems: don’t use CAT5.
So what are the alternatives:
Coax. Twenty years ago network cables consisted of coax cable connected via BNC connectors. It was thick and therefore difficult to install, but it was also much better at shielding noise, as it contained a continuous metal shield around the cables. Unfortunately, this style of cabling was only commonly used on slow ethernet (10Mb/s as compared to modern 1Gb/s - 100 times faster), and finding modern equipment that will use this style of cabling is unlikely.
Part of the reason for there being a lack of modern coax network equipment, is that there is a much better alternative: fibre. Fibre optic cables can support very high transmission speeds and are immune to RF noise interference. However, it is more expensive and harder to install, so tends to be used to connect between networks, or where there is no alternative.
Saying that, fibre optic cable is the best medium for transmitting network signals, and if they could, network engineers would use it everywhere.
Which begs the question: why isn’t hi-fi equipment using fibre optic network connections?
Network switches with fibre optic ports are more expensive than standard switches, but not that much more – an HP E2520-8 8 port switch costs around £260 pounds and has two SFP ports that can take fibre modules (from around £60 each). A fibre network card to fit into a computer can be had for around £70. When you are looking at hi-fi music streamers costing hundreds if not thousands of pounds a go, that’s hardly a lot of money.
So how long will it be until server to network player connections follow the same trend that CD transport to DAC connects went through - from copper to fibre? Surely it has to happen. I only wonder why we haven’t seen it yet!
A StarTech fibre optic PC card available for £96, allows data to be sent down optical fibre.