 
|
Audio Asylum Thread Printer Get a view of an entire thread on one page |
For Sale Ads |
|
|
Audio Asylum Thread Printer Get a view of an entire thread on one page |
For Sale Ads |
In Reply to: I would say all three posted by John Escallier on December 10, 2002 at 14:45:29:
Parallel termination is typically low-impedance driver driving Zo cable to Zo load. There are no reflections from the load back to the driver. I have designed plenty of these types of systems. Point-to-point is the simplest.As I said, a driver impedance of Zo will half the voltage at the load. Very few systems use this technique because it reduces the signal to noise ratio and changes the voltage swing at the load.
In microprocessor buses, such as the front-side-bus in Pentium systems (GTL+), the bus is multi-master, so each driver must have a terminator as well. The drivers are always 7-10 ohms, the bus and the terminators are 50-70 ohms Zo. About 10X difference.
Follow Ups:
Somewhere around 1/10 of a wavelength driving and sink impedences do start to matter.What that amounts to at audio frequencies is pretty boring, but when we're talking about a digital system that must respond to 20 MHz it starts to get a bit more touchy, at least for the cable and termination side.
The driving side may not care about SWR, of course :)
JJ - Philalethist and Annoyer of Bullies
The spectrum of a trapezoidal wave is a series of frequencies extending to GHz.
Thanks for adding something to this, JJ. Last night I got out my TDR meter and made some simple measurements. First, the TDR meter works best at 50 ohms, so I used that as my reference. IF I had a 50 ohm source, 50 ohm cable and a 50 ohm termination, everything was virtually perfect. But IF I added a additional 50 ohm termination at the input of the TDR, to make the effecitve source drive 25 ohms, then I got reflections. IF I used a 75 ohm cable, instead of a 50 ohm cable, then I got reflections, etc etc. This was with 1/2 or 1 meter of cable. Therefore, even a 1/2 meter length of the wrong impedance cable, will change the signal.
It is also important to note that RCA connectors CANNOT be 75 ohms, because the intrinsic ratio between the inner pin diameter and the 'gound' inside diameter of the connector is too small. This has to create reflections or what might effectively look like 'glitches' on the digital waveform.
Ignoring the evidence doesn't do much to improve our understanding of what might cause differences between digital cables.
""It is also important to note that RCA connectors CANNOT be 75 ohms, because the intrinsic ratio between the inner pin diameter and the 'gound' inside diameter of the connector is too small. This has to create reflections or what might effectively look like 'glitches' on the digital waveform.""From that alone, I would worry about the source impedance, as the designed connectors in use will reflect somewhat.
What kind of risetime does your TDR have; risetimes must be a beast, and those kind of slew rates are hard to keep under control, they can tell if your shoes are tied.. Does an RCA connector show as bad a match at the frequencies of the digital audio stream? I imagine the point discontinuity of the RCA would have a frequency dependence component. Perhaps they considered it as a non-issue? Has anybody tried TDR on the standard 75 with RCA's?
Cheers, John
The standard RCA connectors are barely noticed. The 15 nsec risetime is too slow to cause much reflection from a short connector like this. Connector effects are definitely second-order compared to cable losses.
John Es, I use a TEK 1502. It has a 200ps risetime, I think. I do know this: I can easily see the effect of a single RCA-BNC adaptor either at the source or in the line after one meter of cable.
""Parallel termination is typically low-impedance driver driving Zo cable to Zo load. There are no reflections from the load back to the driver.""Ideally, there are no reflections. May not always happen.
""As I said, a driver impedance of Zo will half the voltage at the load. Very few systems use this technique because it reduces the signal to noise ratio and changes the voltage swing at the load.""
I missed where you said that. But I agree.
""In microprocessor buses, such as the front-side-bus in Pentium systems (GTL+), the bus is multi-master, so each driver must have a terminator as well. The drivers are always 7-10 ohms, the bus and the terminators are 50-70 ohms Zo. About 10X difference. ""
They don't have or want the double voltage that would be required with proper driver impedance. And they have stringent requirements on what attaches to the line.
Don't they increase trace width on the board for high fanout runs? I had to do that for ECL hybrids.
And I've always wondered how they get away with hard drive cables with a connector in the middle. Active terminations?
Cheers, John
"They don't have or want the double voltage that would be required with proper driver impedance. And they have stringent requirements on what attaches to the line."I was a member of the group that defined the GTL+ FSB for Pentium for about 8 years. The spec is actually very sloppy in order that motherboards can be implemented in 4-layers (to keep them cheap).
"Don't they increase trace width on the board for high fanout runs? I had to do that for ECL hybrids."
These buses are multimaster, 2 agents in uniprocessor systems (UP) and 5 agents in 4MP (multi-processor) systems. The traces are matched impedance but with the different topologies of the bus they sometimes need to vary the trace width. Certainly in an 4MP 6-layer board they use different width for inner strip-line versus outer micro-strip layers. It is all simulated ad nauseum - sweeping all the variables and then doing monte-carlo runs. Takes months.
And I've always wondered how they get away with hard drive cables with a connector in the middle. Active terminations?
If the ends are terminated properly and the drivers are capable of driving 1/2 Zo, then you can put a load anywhere along the daisy-chained line with no reflections. On the processor bus, the terminations are active.
Diode termination. It is the simplest and produces very clean signals. I use to use it on ECL clock lines, that were used as master clocks and ran a fair distance. The down side of course is power consumption, but if you use it for only a few lines, it shouldn't present a problem.As far as trace width, we always used microstrip or stripline design to control the Z of the trace. This required speccing the distance between layers, the width and weight (height) of the traces, and the prepreg material used between layers. We never used wider traces, just because of a run length.
Of course this was nearly a millennium ago, so design rules for high speed signals may have changed with newer technology. At the time, the highest speed signal we ever ran on a board was 560 MHz (these were for ATE equipment for the semiconductor industry as well as semiconductor research groups.).
However, I don't believe these types of signal termination or propagation issues scale well to digital audio needs.
Tom N.
Many microprocessor buses now use active terminations and diode snubbers to prevent overshoot - usually on-die. The edge-rates are actually faster than those of old ECL - less than 1 nsec. They often have to slow the edges down (to 1-2 nsec) to prevent overshoots which can cause damage to silicon. The difficulty with using fast diodes to absorb overshoots is it can cause large ground-bounces on the die.
Your right Tom, Digital Audio is nowhere near as fast as ECL. The transient problems requiring the diode don't appear to be needed.
With hybrids, 25 mil thick alumina required a 25 mil wide stripline for 50 ohms. Double width, half impedance.And boy, has the technology changed since.
Pre-preg...did a google search on that a while ago for a helical superconductiong magnet design, got Pam Anderson pictures before she was pregnant.
Cheers, John
The key here is point to point. If your not driving point to point then the traditional rules still apply. Point to point, depending on distance, frequency and wavelength can work in the situation you describe. It will not work for longer distances, such as driving a 2 foot cable.
If you are driving point-to-point, the typical parallel termination technique I described will work over long distances, particularly at 3 MHz and 15 nsec risetimes. Should be safe to do 30 feet with no bit-errors using low-loss cable. I designed a supercomputer communication system in the late 80's that did serial 3 Mbits/sec over 20 foot twisted-pair cables and this was parallel terminated using AC termination (resistor in series with capacitor at the receiver).
How does this relate to the SPDIF standard or the AES/EBU standard? It doesn't. SPDIF is a 75 ohm transmission line, and AES/EBU is 110 ohm differential transmission line. For both of the above the output impedance must be 75 ohm for SPDIF and 110 ohm for AES/EBU. Otherwise the reflections will cause timing problems. PERIOD!!!!
< < < < Otherwise the reflections will cause timing problems. PERIOD!!!! > > > >It is not a black and white as you make it. For example, it is virtually impossible to make a cable that is exactly 75 ohms characteristic impedence over its entire length. Therefore, reflections are a given, in all cables.
There are at least two things to look at. First is the length of the cable in relation to the frequency of the signal. In other words, what is the wavelength of the reflection? Transmission line theory only applies when the line is longer than a quarter wavelength.
And even more importantly, the amount of jitter introduced by cable reflections must be correlated to a sonic difference. DACs are now able to handle much more jitter without distortion.
I personally doubt a 3 or 6 foot cable is going to introduce any audible jitter given the 100m wavelength of the S/PDIF signal.
"Transmission line theory only applies when the line is longer than a quarter wavelength."This is an analog rule. For digital, the rule of thumb is more complicated. The rule is generally that if the signal propagates down the line and back and the edge has not finished transitioning yet, then no transmission-line effects will occur - or at least none to worry about.
Good Points. At present the sampling rate of 44.1 kHz, things are moving at a reasonably slow rate so much of your argument makes sense. As the sampling rates increase they won't. I don't agree about your assertion on timing jitter. Unless the unit has a FIFO and reclocks the data,(VCO's are very sesitive to power supply noise) the unit will have sensitivity to jitter. Even Crystal DIR chips aren't perfect.
One of the other problems that we can encounter is that when the SPDIF signals come out of the input connector into the DAC they usually have some amount of PCB etch before it is terminated into 75 ohms. I have my doubts that this anywhere near 75 ohms. This compounds the problem further. When I did my DAC the input SPDIF connector has a 10 Inch run of 75 ohm cable and is wired point to point to the 75 ohm input resistor. Call me crazy, but I have spent a lot of my working life straightening out impedance mismatches.
This is only the aggregate rate of the data part of the frame. The real bit rate is around 3 MHz. There is a ton of non-data stuff in the frames.
Actually, the real bit rate for redbook CD is 705,600 bits per second. The 3 mHz you refer to is the signal rate for the S/PDIF interface where the Manchester endoding uses two clock cycles per bit. The rest, as you have said, is for other, non-audio information.
The real data rate is 44100*16*2=1411200 the manchaster encoding increase the line rate to twice as much pluss the above mentioned overhead. To adequatly transfer the data the cable bandwidth must be about 2.3 times that much. To transfer digital CD quality sound in stereo you need the bandwidth of one analog PAL video channel. It is funny how all those things work out :-). Also, you need one T1 to carry the same raw digital signal over the public network.
dee
;-D
This is just the way that the SP/DIF standard was defined. This is one way to do parallel termination, but it is not the usual way.
This post is made possible by the generous support of people like you and our sponsors:
Top of Page ]
[ Contact Us ]
[ Support/Wish List ]
[ Copyright © 2025, Audio Asylum, all rights reserved ]
