[sdiy] S/PDIF cable. DDD & ADD

[Magnus Danielson]

From: Dave Krooshof <synthos at xs4all.nl>
Subject: Re: [sdiy] S/PDIF cable. DDD & ADD
Date: Sat, 27 Apr 2002 23:01:01 +0200

> Hello Batz,

Hello Batz and Dave,

Interesting to see this S/P-DIF discussion, especially since my
daytime job is to design and standardize telecommunication equipment
and this means digging into obscurities of various systems, including
those we design ourself (uhm, some of it is my design).

> About your DDD post:
> I use video cable for it. That's not RF coax (antenna wire),
> but more like a mono high quality audio cable.
> Works fine over 7 metres in my studio.
> 'Digital cables' is an esotheric subject. Lost of bogus info.
> As spdif is quite high pitched, echo's in the cable can
> confuse the data transport. I can see the cable
> matters then. I guess video cable works, as echo's
> and bad shielding are visible on screen. This will guarantee
> that quality is an issue for every customer.
> Audio cables are only discussed by people with bats ears
> (and a lot of money) who start coax cable hoaxes.

Reflections come up as a result of impedance changes. An important
factor for the actual responce is the risetime of the source. Faults
which is significantly shorter than the rise-time (Say 1/8th) will not
create much of disturbance. Impedance changes will however often do
that. If you only have exactly one impedance error, then you are
mostly safe, you only loose some power, but often you have two gross
onces, and then you are screwed. You get a bunch of reflections and it
takes time for the waveform to "build up" to its correct level since
it takes time before all the reflections have echoed out. One trick is
to lowpass filter the signal, classically to 0.75 of symbolrate with a
4-pole Bessel-Thomson filter. It is also assumed that the signal is
transmitted in about a certain effect range and that the dispersion
increases with how much damped the signal is, so in the input gain
stage you not only raises the gain, but also toss more dispersion
compensation (often refered to equalization). Now, toss this into a
clock recovery PLL. Given that you've done the right thing with cables
and all, the signal dependent part of the jitter and wander will be
kept to a minimum. Excess jitter will cause bit errors, since you will
simply sample the waveform at the wrong point in time. The closer to
the average sampling point you have degraded signal, the more likely
is a signal error. Bit-heads usually take the signal on the scopes and
view the bits, usually of a time scale of 1.5-2 bit large. What you
get is an eye diagram. Between the stable high and low levels there is
a forbidden area which gets properly defined. As long as the signal
doesn't go into that area you usually are quite safe.

There have been concerns that the jitter and wander introduced into
the clock would cause the D/As to phase-modulate the signal. I have
not myself experienced problems with this on audio, so I will not
judge over it, but it has been presented as a possible problem and as
for the technical side of it, I can say that most of audio engineering
of digital transmission is mostly very bad designs, so the problem
could be real.

> It's a one way protocol, so no-one is checking weather data
> was not received well c.q. needs corrections.

All classical tranmission either acheives low bitrates by design
(cable/fibre based - just do it right) or by design (radio based -
just do it right, and toss error correction on it).

Modern transmission systems (SDH, SONET and OTN) do Interleaved
Reed-Solomon (239,255) Error Correction.

Only packet-based stuff actually do retransmission. You could argue
that this is not good engineering (i.e. why loose packets to start
with), but that is another issue.

> I noticed data errors in cables in three ways:
> - a blurred stereo image, this is what causes people to get talked
> into 100$/m cable on hifi lists. ("oh, your using *that* kind of
> cable. Well, that's just like flushing all the money you spend
> on that seperate DAC down the drain.")
> In fact, it might be the error correction doen't know how
> to correct some of the wrong bits.

Sorry? Error correction?

There is none! Just for the sake of it I just checked with the IEC
958, and found none. I've been digging around this stuff for quite
some time and never heard of error correction. If you get a bit error,
you get it full-blown. There is no error correction to even see it.

If fact, there is no CRC check for the validity of the audio-content,
only a CRC for the channel status. There is a parity bit for the
frame, but that's about it.

The only error correction we talk about is in the CD-player.

> - a sharp sound, due to the precence of errors in bits
> that were sort of corrected.
> - ticks. errors that cause the wordlength to be misinterpreted.
> 
> All these were corrected in using just any good video cable
> with proper shielding, nicely soldered plugs, and bought
> during new moon.

There's no magic to it. Good impedance matching, properly done
cables and well designed inputs and outputs contribute to a minimum of
bit errors and jitter/wander.

Just for fun, I picked up a random 3 meter microphone cable and let my
Network analyser have a go at it. I had put XLR contacts on a pair of
coaxes and normalized the responce with that. Anyone care to guess
which bandwidth I had, i.e. at which frequency did I first drop below
-3 dB?

Cheers,
Magnus