And how do you time the reflections...
Add a terminating resistor (parallel at the receiver) to absorb the excess energy (the overshoot and ringing) as ringing stresses the silicon, or change the transmitters drive strength (series resistor at the source).
Have a look at this:
Proper Termination for High-Speed Digital I/O Applications - National Instruments
Further look up the PCB Layout 6" rule or 1ns rule...
For a 1ns rise time use a trace length of 3" or less, then any reflections will arrive after the signal has reached the destination.
Then look at electrical length and the signals rise times...
http://www.analog.com/static/imported-files/tutorials/MT-097.pdf
Twisted pair go to differential pair routing. The impedance is controlled by trace width, distance from ground plane, microstrip or stripline etc, use one of the attached calculators to work out the required dimensions. USB and other interfaces you can also find good guidelines in driver and receiver chips data sheets.
Have a look at, all require at least one ground plane (as digital does anyway):
PCB Impedance and Capacitance Calculator
Missouri S&T index
Saturn PCB Design - PCB Via Current | PCB Trace Width | Differential Pair Calculator | PCB Impedance
And for the really serious:
PCB signal integrity and controlled impedance – PCB troubleshooting
If you read and understood the article you would know that is not what he actually said. He actually said that you want to avoid having the first reflection from the beginning of the transition arrive in the middle of the transition. When that happens depends on the exact shape and rise time, and the cable velocity factor. Nothing magical about 1.5m, even if his idea is correct.thetubeguy1954 said:
The sharpest part of a transition is often the middle, so that is the part which is most likely to cause troublesome reflections. The beginning and end are often smoother. There may be an argument in avoiding a transition reflection hitting the next transition, but that just sets a much longer cable length as one to avoid and it will depend on sampling rate.
There may be an argument for longish, lower quality cables being best as they will attenuate reflections more than a short high quality cable. But there: I am straining at gnats again.
Having patents (especially in the US) tells you precisely nothing about the skills of someone, unless you have read and understood their patents and can appreciate their creativity.
Placebo effect
One of the two SPDIF sources is a PLEXTOR CD ( PC ).
I have changed the Clock and the SPDIF is reclocked .The SPDIF signal is drived by a RS485 driver.....
This source is better than a commercial CD reader
Serge
Perhaps I should have added 'user-modified equipment' to my list of possible problems. I know enough not to dabble - I hope that whoever designed the circuit and PCB knows more about mixed-signal design and low noise oscillators than I do. In many of these areas I know enough to be aware of the problems but not enough to arrive at reliable solutions. People knowing a lot less or a lot more than me don't have this handicap.
What about the reflections interacting with the other bits that follow? if one is so concerned about how much effect this is going to have then you are better getting a scope, Howard Johnsons book of black magic and some resistors...and terminating the line (from transmitter to receiver, including the cable) properly. The excess energy that is being reflected also stresses the silicon connected to the line.
This on reflection is probably to avoid non-monotonic rising edges, but as DF96 has said, how do you tune your cable length to the actual properties of your system. You need to know the drive strength of the transmitter (this info can be found in the IBIS data for the device, if you can get it), the impedances of the lines, the dielectric constant of the cable, this will give you the Vf of the signal so you can work out how fast its gonna travel, so you can work out the electrical lengths involved.
Now as I have stated a nice gentle digital waveform with slightly rounded corners is best, some minor overshoot and ringing is tolerable, but not really desired, but non-monotonic rising (or falling) edges are the killer, these can lead to false switching and proper bugger things up. Some more pretty picture etc:
Printed Circuit Design & Fab Magazine Online
A bit more info for AndrewT.
SPDIF is single ended so you calculate for single ended routes.
USB is differential, so calculate for diff pair (usually 2X Single ended), USB is also 90r, whereas ethel-net is 100r, so on boards with both the USB usually ends up 100r also.
Ground planes are critical, and looking at some DACs and sources I despair at the cheapness of the PCBs (single sided!).
SPDIF is single ended so you calculate for single ended routes.
USB is differential, so calculate for diff pair (usually 2X Single ended), USB is also 90r, whereas ethel-net is 100r, so on boards with both the USB usually ends up 100r also.
Ground planes are critical, and looking at some DACs and sources I despair at the cheapness of the PCBs (single sided!).
Note that the edge rate determines the smallest feature size for which impedance matching is significant, spdif symbol rate is a few Mhz, so if you allow an edge rate of say 100Mhz that should cover even very stiff drivers.
At 100Mhz, in a cable with 66% Vf (Typical of solid dielectric coax) a wavelength is 2M, so allowing that nothing smaller then about 1/10th of a wavelength is significant in a transmission line sense, connectors and board traces shorter then 20cm or so can be treated as lumped constant networks, which is to say pretty much ignored in terms of impedance matching.
If you have to run a significant (in a transmission line sense) length of line to a misterminated load, one can reduce the ISI by source terminating the line at the cost of 6dB less level if the line does end up being correctly terminated. This series termination absorbs the reflected power at the source so it is not re reflected to cause interference at the load.
This observation is not to be taken to imply that I think ISI is any sort of issue on an spdif link (That fails the giggle test).
If basically competent wires sound different with a given set of gear when used for spdif, I would take that to mean that the equipment in question should be replaced by stuff that works properly, because the clock recovery is clearly sub par on the stuff you are using at the moment.
Regards, Dan.
At 100Mhz, in a cable with 66% Vf (Typical of solid dielectric coax) a wavelength is 2M, so allowing that nothing smaller then about 1/10th of a wavelength is significant in a transmission line sense, connectors and board traces shorter then 20cm or so can be treated as lumped constant networks, which is to say pretty much ignored in terms of impedance matching.
If you have to run a significant (in a transmission line sense) length of line to a misterminated load, one can reduce the ISI by source terminating the line at the cost of 6dB less level if the line does end up being correctly terminated. This series termination absorbs the reflected power at the source so it is not re reflected to cause interference at the load.
This observation is not to be taken to imply that I think ISI is any sort of issue on an spdif link (That fails the giggle test).
If basically competent wires sound different with a given set of gear when used for spdif, I would take that to mean that the equipment in question should be replaced by stuff that works properly, because the clock recovery is clearly sub par on the stuff you are using at the moment.
Regards, Dan.
Not true at all. The circuitboard should provide a very low impedance. At the output connection point, a 75 ohm resistor is wired in series with the feed to the center conductor. This is what produces a true 75 ohm output impedance. It's the same design which has been used with analog video for many decades.
At the receiving end of the coax, there should be a 75 ohm termination.
Following these guidelines will prevent cable echo problems.
In any case, a "digital cable" in which form is different from an analog one?
Characteristic impedance. Not critical for analog, somewhat critical for short runs of digital, very critical for long runs of digital.
Then RG8, RG59, RG213 and others, cannot work in digital environments as they has been designed to conduct analog signals from Tx's /Rx's and antennas.
What I means is that a cable cannot distinguish from digital and analog signals, it's a simple piece of copper. So no "digital cable" or "analog cable", it would be said cable for digital communications. The cable itself isn't "digital".
What I means is that a cable cannot distinguish from digital and analog signals, it's a simple piece of copper. So no "digital cable" or "analog cable", it would be said cable for digital communications. The cable itself isn't "digital".
Why not, if the characteristic impedance is correct? RG8 and RG213 are not correct for digital audio standards- they're 52 ohms.
Yes, absolutely.
It is really sad to see how many well respected audio components get the basics wrong, all the while claiming some special knowlege that somehow every other mixed signal man on the planet (Some dealing with things a deal MORE critical then mere audio) has missed.
I think it is because audio (as opposed to digital stuff) is always so slow that all sorts of poor layout can sort of work, and things like star earthing seem attractive when you are going so slow that lead inductances are unimportant.
Any time you see modern digital or mixed signal on a two layer board, they are doing it wrong (or at least being forced into a heavy BOM cost compromise, 4 layers is not ideal, but at least you have a chance to get a solid plane in there, also track sizes for sane impedances become more reasonable)!
Any time you see a split in a ground plane under a mixed signal part it was probably put there by a designer who was 'hard of thinking' (Yes I have seen those app notes as well), very occasionally it is the right thing, but so very seldom.
Star 'earths' are a horrible plan in high speed mixed signal designs, like say dacs, use a solid plane instead and be selective about where you route things.
When considering a board, think always about current loops, and remember that above a few hundred khz, inductance usually dominates and return currents will really want to flow directly under the trace carrying the outgoing current in order to minimise the energy in the magnetic field.
This effect is strong enough that there is usually no margin in splitting ground planes because the current will naturally flow under the conductor that forms the other part of the loop without any help from you.
Remember that a signal trace on the outside of a board will be referenced to the plane immediately below, so using a via to take it through the board also requires tiying the two planes together (at least at AC) at that point, particularly with 4 layer stackups with buried power and ground plains (the usual plan) you should place a decoupling cap between the planes beside any via that takes a high frequency trace from one side to the other.
All of this looks suspiciously like capital E Engineering however, so folks would rather talk about transmission line effects in cables clearly too short to exhibit them, and argue about the virues of BNC vs Phono vs PL259 vs Lemo vs N type vs 7/16th connectors for a barely RF signal...
It is depressing.
Regards, Dan.
I think it is because audio (as opposed to digital stuff) is always so slow that all sorts of poor layout can sort of work, and things like star earthing seem attractive when you are going so slow that lead inductances are unimportant.
Any time you see modern digital or mixed signal on a two layer board, they are doing it wrong (or at least being forced into a heavy BOM cost compromise, 4 layers is not ideal, but at least you have a chance to get a solid plane in there, also track sizes for sane impedances become more reasonable)!
Any time you see a split in a ground plane under a mixed signal part it was probably put there by a designer who was 'hard of thinking' (Yes I have seen those app notes as well), very occasionally it is the right thing, but so very seldom.
Star 'earths' are a horrible plan in high speed mixed signal designs, like say dacs, use a solid plane instead and be selective about where you route things.
When considering a board, think always about current loops, and remember that above a few hundred khz, inductance usually dominates and return currents will really want to flow directly under the trace carrying the outgoing current in order to minimise the energy in the magnetic field.
This effect is strong enough that there is usually no margin in splitting ground planes because the current will naturally flow under the conductor that forms the other part of the loop without any help from you.
Remember that a signal trace on the outside of a board will be referenced to the plane immediately below, so using a via to take it through the board also requires tiying the two planes together (at least at AC) at that point, particularly with 4 layer stackups with buried power and ground plains (the usual plan) you should place a decoupling cap between the planes beside any via that takes a high frequency trace from one side to the other.
All of this looks suspiciously like capital E Engineering however, so folks would rather talk about transmission line effects in cables clearly too short to exhibit them, and argue about the virues of BNC vs Phono vs PL259 vs Lemo vs N type vs 7/16th connectors for a barely RF signal...
It is depressing.
Regards, Dan.
It is indeed, but unfortunately, there's a vested interest in promoting ignorance and misinformation. As DF96 likes to say, the issue is at the very lowest and very highest end of the market.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.