I'm doing oversampling in a DSP and I want to provide some isolation between it and the DAC chip, so I'm thinking something like this:
DSP-->pulse transformer-->reclocking-->DAC
For the transformer, I think a regular Ethernet tranfsormer would be fine. What would be the best option for the reclocking? The signal would be 96 kHz 24 bit oversampled 4x. I'm going to use a clock < 3 ps jitter (starting 10 Hz offset from the fundamental) so I want to get the most out of the ultra low jitter clock.
DSP-->pulse transformer-->reclocking-->DAC
For the transformer, I think a regular Ethernet tranfsormer would be fine. What would be the best option for the reclocking? The signal would be 96 kHz 24 bit oversampled 4x. I'm going to use a clock < 3 ps jitter (starting 10 Hz offset from the fundamental) so I want to get the most out of the ultra low jitter clock.
Well you can steal from my project :
http://www.diyaudio.com/forums/showthread.php?postid=1398512#post1398512
You want to separate your stuff in 2 halves with isolators. The Clock should go into the same half as the reclocking+DAC. The DSP should go in the other half.
This removes the problem of transmitting a low jitter clock through the isolator.
Also the clock must not use a noisy power supply contaminated by the DSP.
Now for isolation, for a clock signal (pure AC), a transformer is perfectly fine, and 100 Mbps Ethernet transformer should fit the bill. Get the part number from my schematics (caveat : it's not built yet but the specs say it should work).
However data, which is NOT pure AC, will NOT go through a transformer, don't even try. DC will saturate the core and all you'll get will be random bits.
You need either :
- a transformer with some sort of biphase coding to make the waveform AC (complicated)
- same as above, integrated in a chip ready to use (AD makes those, it's ADUMxxxx I believe, not sure)
- or a fast opto
Also you could simply use optical fiber with TORX/TOTX for both clock and data ; you need to get high throughput parts for the clock though, check the datasheet and be sure to remember that the throughputs are specified in Mbps which means a clock period counts as 2 bits (high+low) so you have to divide by 2 to get the highest clock freq those will pass.
http://www.diyaudio.com/forums/showthread.php?postid=1398512#post1398512
You want to separate your stuff in 2 halves with isolators. The Clock should go into the same half as the reclocking+DAC. The DSP should go in the other half.
This removes the problem of transmitting a low jitter clock through the isolator.
Also the clock must not use a noisy power supply contaminated by the DSP.
Now for isolation, for a clock signal (pure AC), a transformer is perfectly fine, and 100 Mbps Ethernet transformer should fit the bill. Get the part number from my schematics (caveat : it's not built yet but the specs say it should work).
However data, which is NOT pure AC, will NOT go through a transformer, don't even try. DC will saturate the core and all you'll get will be random bits.
You need either :
- a transformer with some sort of biphase coding to make the waveform AC (complicated)
- same as above, integrated in a chip ready to use (AD makes those, it's ADUMxxxx I believe, not sure)
- or a fast opto
Also you could simply use optical fiber with TORX/TOTX for both clock and data ; you need to get high throughput parts for the clock though, check the datasheet and be sure to remember that the throughputs are specified in Mbps which means a clock period counts as 2 bits (high+low) so you have to divide by 2 to get the highest clock freq those will pass.
peufeu said:
Yes, that was my intention from the beginning, and I'll even have the two halves separately shielded.
Now for isolation, for a clock signal (pure AC), a transformer is perfectly fine, and 100 Mbps Ethernet transformer should fit the bill. Get the part number from my schematics (caveat : it's not built yet but the specs say it should work).
However data, which is NOT pure AC, will NOT go through a transformer, don't even try. DC will saturate the core and all you'll get will be random bits.
You need either :
Well, the DSP can do the coding on the transmitting side, so all that remains is the decoding. But thanks for mentioning the ADUM parts; that's probably the best solution (I'm not sure really whether optical or the biphase coding and transformer have any advantages over each other).
My main question is about the choice of reclocking ICs. In your schematics you have SN74HC74D, so I'm wondering how you came to that choice. These max out at 25 MHz so in my case where the audio is oversampled, it's pushing it.
What about using something like 74ACTQ273 or 74ACTQ574? Though these are octal and I didn't find dual or pico-gate analogues.
Oh, by the way, I'm Drawknob at the other place if you didn't figure it out LOL
Actually Tent (aka The Man) advised to use HC and specifically not HCT. Apparently HCT has higher jitter. This is probably due to the TTL instead of CMOS levels (CMOS level switch is close to Vcc/2 which places it in the area where the signal has highest slew rate, which is optimal). However I guess you could use AC logic if you want more speed... those have very fast edges, though, so watch out for transmission line effects. But you're probably using source termination resistors anyway ?...
Data and bit rate is over 9 MHz because it's oversampled by an external DSP. Yes, I intend to use termination resistors. Would ACT and ACTQ be equivalently as good, by the way? To what extent the "quiet" part matters in this application I'm not sure.
As for picogates, it looks good because of more isolation between lines, and one could match for propagation delay, but it's a new type of part for me and I don't know how to choose...
As for picogates, it looks good because of more isolation between lines, and one could match for propagation delay, but it's a new type of part for me and I don't know how to choose...
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.