Thanks - any idea where to get 50R coax? I can't find any on Partsconnexion. The connection is barely 10cm, so ribbon cable is an option, if I can find some with 6 wires. I assume you mean the ribbon cable where the casing is all connected?
Why does the signal need to be close to the ground companion?
if it doesnt have a nearby ground/return for every signal, it will affect the signal integrity, but also spray RF around your case. i2s signals, particularly MCLK are quite high frequency, they are up in the radio frequency bands. high frequencies/radio frequencies travel quite well through the air, as you know, because the air presents a low impedance at high frequencies, so if there is no lower impedance path nearby, it will just find its own way through the air to the nearest ground or signal trace or component in a fairly unpredictable way. return signals even on a ground plane, will tend to travel as close to the signal trace on the other side because naturally it prefers to be as close as possible.
i'll have a look if I have some spare, or alternatively you can buy some at element14. I wouldnt use ribbon cable really, mini coax is preferred
i'll have a look if I have some spare, or alternatively you can buy some at element14. I wouldnt use ribbon cable really, mini coax is preferred
Last edited:
qusp is generally correct. Though I do have to make a correction, the current flow is not via the air in this case. That would mean that the space between the signal and return conductor/pcb are ionised. Which I am reasonably sure isn't going to happen with the signals we're dealing with in I2S. I am certain he understands the issue (I've talked with him in person about it a few times already) most likely that he was in a hurry when he typed that out.
I felt across the forum this question comes up often enough that its worth ensuring it is written out somewhere.
Here is my take on behaviour of I2S signals and how the diy-er should treat them:
Consider that between the signal and return conductors there exists a magnetic field as well as electric field. An increased area between them decreases their efficiency and more of the magnetic and electric field radiates rather than being canceled out by the opposing field generated in the return conductor. This is to say that the signal behaves less like an electric transmission line; it starts to behave more as an RF transmission antenna.
With high frequency signals, a poorly terminated electrical transmission line will radiate out the 'open' end and have some portion of the signal reflected back down the transmission line. If the 'loop area' is large there will be radiation generated along the transmission line.
This comes down to two things:
This is to ensure both the signal sent and received correlate closely AND to minimise the radiated noise that will contribute to the noise in components nearby.
The distance at which the conductor separation becomes significant depends on the frequency of the signal (ie the combined harmonics needed to produce the 'square' wave). Refer here - Red Flags -I read that to mean that the distance between signal and its return conductor should be less than 1/20th of the highest harmonic needed to form the 'square' wave. Note that the 'squarer' the analogue waveform, the higher the number of harmonics required to produce that wave.
This is why the signal transmission is considered an analogue system and termination resistors help slow the rise time to reduce the band of frequencies required to produce the signal.
In the end, twisted pairs of copper may work quite well if the pcb has proper termination resistors for each signal AND the sending and receiving PCB have a nearby ground termination point at both ends.
For absolute best signal termination a signal will travel a pcb trace with a characteristic impedance matched to a termination resistor, leave the pcb and travel an impedance matched coax cable, land and the next pcb and have a pcb with similarly well designed pcb trace and termination resistor. For best performance of the termination of the cable the signal landing pad will have a circular return path termination around it. As the size of this is reduced, the performance increases. This is because it reduces the 'opening' at the end of the transmission line.
And this boys and girls, is why the smart guys on these pages use u.fl and w.fl terminations on all of their high speed signals. These coaxial cables tick all of the boxes, all you need to do is preach the good word to the designers for the devices at either end of the cable to keep up their end of the bargain; or speak with your wallet and support the guys who are 'doing it right'.
Chris
I felt across the forum this question comes up often enough that its worth ensuring it is written out somewhere.
Here is my take on behaviour of I2S signals and how the diy-er should treat them:
Consider that between the signal and return conductors there exists a magnetic field as well as electric field. An increased area between them decreases their efficiency and more of the magnetic and electric field radiates rather than being canceled out by the opposing field generated in the return conductor. This is to say that the signal behaves less like an electric transmission line; it starts to behave more as an RF transmission antenna.
With high frequency signals, a poorly terminated electrical transmission line will radiate out the 'open' end and have some portion of the signal reflected back down the transmission line. If the 'loop area' is large there will be radiation generated along the transmission line.
This comes down to two things:
- Each signal must have independent return path, so that each signal acts as an independent transmission path.
- The return path should be as close as practical to the signal.
This is to ensure both the signal sent and received correlate closely AND to minimise the radiated noise that will contribute to the noise in components nearby.
The distance at which the conductor separation becomes significant depends on the frequency of the signal (ie the combined harmonics needed to produce the 'square' wave). Refer here - Red Flags -I read that to mean that the distance between signal and its return conductor should be less than 1/20th of the highest harmonic needed to form the 'square' wave. Note that the 'squarer' the analogue waveform, the higher the number of harmonics required to produce that wave.
This is why the signal transmission is considered an analogue system and termination resistors help slow the rise time to reduce the band of frequencies required to produce the signal.
In the end, twisted pairs of copper may work quite well if the pcb has proper termination resistors for each signal AND the sending and receiving PCB have a nearby ground termination point at both ends.
For absolute best signal termination a signal will travel a pcb trace with a characteristic impedance matched to a termination resistor, leave the pcb and travel an impedance matched coax cable, land and the next pcb and have a pcb with similarly well designed pcb trace and termination resistor. For best performance of the termination of the cable the signal landing pad will have a circular return path termination around it. As the size of this is reduced, the performance increases. This is because it reduces the 'opening' at the end of the transmission line.
And this boys and girls, is why the smart guys on these pages use u.fl and w.fl terminations on all of their high speed signals. These coaxial cables tick all of the boxes, all you need to do is preach the good word to the designers for the devices at either end of the cable to keep up their end of the bargain; or speak with your wallet and support the guys who are 'doing it right'.
Chris
Reflections
I think I missed some info on signal reflections.
For these I imagine my high school science light experiments. As light travels from air->glass, at the interface between these two some light is reflected back and some continues. This is exactly the same in this case. Think of electrical signal as your light and characteristic impedance as your optical density.
For this reason I believe that there is a disadvantage in soldered wire terminations where the impedance at the interface is not controlled.
I think I missed some info on signal reflections.
For these I imagine my high school science light experiments. As light travels from air->glass, at the interface between these two some light is reflected back and some continues. This is exactly the same in this case. Think of electrical signal as your light and characteristic impedance as your optical density.
For this reason I believe that there is a disadvantage in soldered wire terminations where the impedance at the interface is not controlled.
As signal frequency increases the return current will follow more closly the path of least inductance, on a PCB it will track under the trace on the ground or power plane(that is hopefully adjacent to the tracking layers)on cables use either a co-ax or wo layer flexi cables with a contigous grouns plane.
The signals travel as a psuedo TEM wave or TEM wave dependant on the cable configuration or whether the trace is routed as microstrip (top or bottom layer, adjacent ground) or stripline(between two grounds). Any seperation between a signal and its return path will cause signalintegriy problems and increase the risk of EMC problems such as radiating noise (which is what Qusp was reffering to, not ionisation.) For ultimate signal integrity a signal should be tracked next to an adjacent ground plane so the e and h field terminate at this ground plane. The frequency spectrum of a square wave that we are concerned about is determined from:
Fknee=0.5/Tr, where Tr is the 10-90% risetime of the signal, this also determines whether a design is high speed or not.
The best sources of info for thie are Henry Ott for the EMC implications:
home page
And either of these two for signalintegrity:
beTheSignal.com
Signal Consulting, Inc. - Dr. Howard Johnson
A couple of links that help illustrate the points.
http://www.ultracad.com/articles/slots.pdf
Printed Circuit Design & Fab Magazine Online
The signals travel as a psuedo TEM wave or TEM wave dependant on the cable configuration or whether the trace is routed as microstrip (top or bottom layer, adjacent ground) or stripline(between two grounds). Any seperation between a signal and its return path will cause signalintegriy problems and increase the risk of EMC problems such as radiating noise (which is what Qusp was reffering to, not ionisation.) For ultimate signal integrity a signal should be tracked next to an adjacent ground plane so the e and h field terminate at this ground plane. The frequency spectrum of a square wave that we are concerned about is determined from:
Fknee=0.5/Tr, where Tr is the 10-90% risetime of the signal, this also determines whether a design is high speed or not.
The best sources of info for thie are Henry Ott for the EMC implications:
home page
And either of these two for signalintegrity:
beTheSignal.com
Signal Consulting, Inc. - Dr. Howard Johnson
A couple of links that help illustrate the points.
http://www.ultracad.com/articles/slots.pdf
Printed Circuit Design & Fab Magazine Online
Signal reflections are caused by impedance mismatches, and are treated by:
Reducing the signal Tr, thus reducing the current drive of the ransmitter, or by either a combination of or one off series and parallel termination. Effectively a reflection is due to to much energy being injected into the line and the reciever being unable to sink all the current.
Reducing the signal Tr, thus reducing the current drive of the ransmitter, or by either a combination of or one off series and parallel termination. Effectively a reflection is due to to much energy being injected into the line and the reciever being unable to sink all the current.
no I meant what I said, you just misread me, perhaps I needed a paragraph break in there between sentences to reinforce I was talking about 2 different forces. I dont mean the return current travels through the air, just that noise resulting from the HF signal can easily become airbourne if you dont have a return path nearby. these are radiated noise currents (I guess they are still called currents?) generated by the HF and high slewrate signal interacting with the surrounding components via electromagnetic coupling
marce explained it better than me of course, but I wasnt saying that the signal would spontaneously leave the wire, not even quite sure what you are thinking I was trying to say TBH =) but I freely admit this whole subatomic wavelet soup hurts my head to think about.
Last edited:
A Google search for 'I2S impedance' gives no useful results on the first 6 pages, apart from hints that no impedance is defined. My guess is that it is a short range digital interface spec for use on one PCB, with the actual impedance being a matter for the designer. Most digital interfaces can cope with some back-termination to damp reflections. My guess is that 50 ohms might be a bit too low, but I am sure marce can tell us what a 'typical' track impedance might be.
If I really needed to get I2S from one board to another and could not put them on the same board then I would use either ribbon (with alternate conductors grounded) or twisted pair. Coax probably introduces too much capacitance/too low impedance for a normal digital output port.
If I really needed to get I2S from one board to another and could not put them on the same board then I would use either ribbon (with alternate conductors grounded) or twisted pair. Coax probably introduces too much capacitance/too low impedance for a normal digital output port.
Any 50 ohm coax has about the same capacitance; the physics requires this. A very low capacitance coax is not 50 ohm, but higher impedance. I doubt if I2S requires microwave quality cables, as it was intended for normal PCB traces. Ribbon is fine for short distances, if properly driven and terminated.
OK, you know best.... probably right on all being the same, but what I do know is u.fl is better than ribbon, this has been confirmed by Ian (designed the best i2s interfacve i'm aware of for this place and works in medical tech, Acko (who works in Mil, RF and comms as well as audio) and marce who I gather deals with transmission lines quite a lot, yet hates ribbon.
u.fl connectors and cables are cheap as if you know where to look, so I cant imagine the cable is pricey, cost you ~15 to sort out sockets and prefabbed cables for 4 wire i2s connection (including mclk). a i2s line between boards without a buffer would be a bit silly anyway
its probably more about the soldering or connectors being bad for i2s, i2s runs are pretty massive speeds these days.
ribbon will 'work' of course, but does short runs really have much to do with it? wouldnt have thought so. ZIF are decent too, the double sided stripline ones, but thats not going to be useful here. I use belden twisted ptfe/copper pair sometimes too and it works well, but thats 100ohms differential, so would depend on the driver and termination.
u.fl connectors and cables are cheap as if you know where to look, so I cant imagine the cable is pricey, cost you ~15 to sort out sockets and prefabbed cables for 4 wire i2s connection (including mclk). a i2s line between boards without a buffer would be a bit silly anyway
its probably more about the soldering or connectors being bad for i2s, i2s runs are pretty massive speeds these days.
ribbon will 'work' of course, but does short runs really have much to do with it? wouldnt have thought so. ZIF are decent too, the double sided stripline ones, but thats not going to be useful here. I use belden twisted ptfe/copper pair sometimes too and it works well, but thats 100ohms differential, so would depend on the driver and termination.
Last edited:
Bear in mind that things like line impedance are transmission line effects and only become significant once the edge rate becomes high enough to put meaningful energy at wavelengths shorter then about 8 times the length of the connection.
Thus the discontinuity presented by a solder joint or by using a 50ohm BNC in a 75 ohm system, or by using a RCA plug for spdif is very unlikely to matter for digital audio (The electrical length is too short).
When hooking up fast signals, a useful guide is to remember that current at any real frequency always tries to return by the path of lowest impedance, and as inductive effects can easily dominate, that usually means 'by the path that minimises loop area', the smalller the loop areas the less energy is stored in the magnetic field and (usually) the better everything behaves.
For a hookup like the OP describes, I would probably go with a ribbon cable with every other conductor grounded at both ends, and maybe with 22ohm or so series resistors right at the transmitter to damp the lines, odds are it will be fine.
Coax for this is iffy, I2C does not usually have a well defined impedance, and usually expects to see a full level swing at the recever, both of which argue aagainst a transmission line treatment.
Regards, Dan.
Thus the discontinuity presented by a solder joint or by using a 50ohm BNC in a 75 ohm system, or by using a RCA plug for spdif is very unlikely to matter for digital audio (The electrical length is too short).
When hooking up fast signals, a useful guide is to remember that current at any real frequency always tries to return by the path of lowest impedance, and as inductive effects can easily dominate, that usually means 'by the path that minimises loop area', the smalller the loop areas the less energy is stored in the magnetic field and (usually) the better everything behaves.
For a hookup like the OP describes, I would probably go with a ribbon cable with every other conductor grounded at both ends, and maybe with 22ohm or so series resistors right at the transmitter to damp the lines, odds are it will be fine.
Coax for this is iffy, I2C does not usually have a well defined impedance, and usually expects to see a full level swing at the recever, both of which argue aagainst a transmission line treatment.
Regards, Dan.
yeah well with dacs pushing clock speeds of 100MHz in todays dacs, and the use of fast logic, I tend to worry about this sort of stuff. u.fl and w.fl connectors and cables have proven to be more reliable with lock than ribbon and lowest DPLL bandwidth on my dac and others using the same equipment.
thanks for the detailed reply though
bare in mind i'm not dealing with a fuzzily defined system here, all points these connectors and cables have been used the termination, impedance and frequency is known and consistent
thanks for the detailed reply though
bare in mind i'm not dealing with a fuzzily defined system here, all points these connectors and cables have been used the termination, impedance and frequency is known and consistent
Last edited:
Sure, even if MCLK is running 25Mhz or so, and has an edge rate 20 times that, the wavelength of the 20th harmonic (500Mhz) is still the thick end of a meter at 0.6 vf, so 10cm or so of cable is a non issue.
Regards, Dan.
You want fun and games, try a 16 bit, 250MS/s ADC, worse try running several of them in sync...
8 Layer board, just so everything can run stripline for the inputs, clocks and the (lots of) LVDS pairs to the FPGA, with low single digit ns cycle times and hundred ps edge rates, fun ****.
You learn a LOT about good layout, and I am certainally learning some painful lessons about signal quality, clock skew and ground bounce, the RF parts were easy, getting the power decoupling, clock skew and signal quality issues sorted is kicking my ****.
Regards, Dan.
8 Layer board, just so everything can run stripline for the inputs, clocks and the (lots of) LVDS pairs to the FPGA, with low single digit ns cycle times and hundred ps edge rates, fun ****.
You learn a LOT about good layout, and I am certainally learning some painful lessons about signal quality, clock skew and ground bounce, the RF parts were easy, getting the power decoupling, clock skew and signal quality issues sorted is kicking my ****.
Regards, Dan.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.