After a long delay I am finally starting to work on my dac project again. Many of the stated goals are quite similar to those found in the Opus thread with the exception that I wanted current output dacs and chose the PCM1794/1798 series.
This is my first ground up dac design so I expect there is quite a lot I don't know or may possibly have missed.
I was unable to acquire PCM1794 at the time I was starting down this road so what I have are a couple of 1798. They are a few dB noisier IIRC, filter slopes may also be different, but I believe the devices are relatively interchangeable should I want to upgrade later.
I have not started to do the pcb design yet, but will be using ExpressPCB to do the layout. (I've used it a lot and the boards are well made and fairly cheap.)
I am probably going to use transformers from either Sowter or Llundahl for I/V conversion. Some sort of tube I/V is a possibility as well, although based on recent experience the transformer I/V seems more transparent.
The supply is on the board.
Currently I have made provisions for I2S mode operation and LJF.
I have provided probably an unnecessarily complex range of configuration choices, but not having used either of these chips in a design before, and having reviewed the design guidelines, specs, and demo board info I have decided to play it safe. Jumpers are cheap, and hacked/trashed pcbs aren't necessarily..
Anyone have experience with this dac and receiver?
Specifically: dac filter rate - seems like I want the sharp roll-off filter due to better in band flatness and bandwidth. Why would I want the other? Seems to roll off quite early, better phase response in band??
Also in the CS8416 there are two options for phase detector update rate. I made them jumper selectable. Could not find a specific reference to why this option is necessary.
I don't currently have plans to offer this to the world at large, but if anyone is interested just let me know.
I will start the layout shortly and what I know from prior digital experience is there will be just one main groundplane, (any analog planes just local to the chip they are associated with) beyond that I have not decided.
Here is the schematic.
This is my first ground up dac design so I expect there is quite a lot I don't know or may possibly have missed.
I was unable to acquire PCM1794 at the time I was starting down this road so what I have are a couple of 1798. They are a few dB noisier IIRC, filter slopes may also be different, but I believe the devices are relatively interchangeable should I want to upgrade later.
I have not started to do the pcb design yet, but will be using ExpressPCB to do the layout. (I've used it a lot and the boards are well made and fairly cheap.)
I am probably going to use transformers from either Sowter or Llundahl for I/V conversion. Some sort of tube I/V is a possibility as well, although based on recent experience the transformer I/V seems more transparent.
The supply is on the board.
Currently I have made provisions for I2S mode operation and LJF.
I have provided probably an unnecessarily complex range of configuration choices, but not having used either of these chips in a design before, and having reviewed the design guidelines, specs, and demo board info I have decided to play it safe. Jumpers are cheap, and hacked/trashed pcbs aren't necessarily..
Anyone have experience with this dac and receiver?
Specifically: dac filter rate - seems like I want the sharp roll-off filter due to better in band flatness and bandwidth. Why would I want the other? Seems to roll off quite early, better phase response in band??
Also in the CS8416 there are two options for phase detector update rate. I made them jumper selectable. Could not find a specific reference to why this option is necessary.
I don't currently have plans to offer this to the world at large, but if anyone is interested just let me know.
I will start the layout shortly and what I know from prior digital experience is there will be just one main groundplane, (any analog planes just local to the chip they are associated with) beyond that I have not decided.
Here is the schematic.
kevin,
I've been working on a similar design. The boards are made and are going to be tested very soon. If you're interested in following the progress of my project (CS8416->AD1896->PCM1794->passive (resistor) I/V+op amp gain), just check out my (non-commercial) website.
Edit: I should add that I used ExpressPCB, and found it convenient, especially because I ordered the mini-board prototype.
I've been working on a similar design. The boards are made and are going to be tested very soon. If you're interested in following the progress of my project (CS8416->AD1896->PCM1794->passive (resistor) I/V+op amp gain), just check out my (non-commercial) website.
Edit: I should add that I used ExpressPCB, and found it convenient, especially because I ordered the mini-board prototype.
Here is the latest version of the schematic. I have started very tentatively to layout the pcb and I am attempting to fit it all onto the standard ExpressPCB protoboard in order to minimize board cost.
Unfortunately that means that some things don't seem to fit too well.
I have eliminated the rectifier, cm choke and large filter cap to save space. I will probably use a cm choke in the raw supply which will be located elsewhere.
I will place all of the voltage regulators which are now all sot-223 sized parts on the back side of the board along with the supply caps.
I have eliminated the external I2S connector as I will never use it, and at higher data rates it represents an unterminated stub.
Changed to the NDS7002 fet (sot-23 package) for the mute led as the threshold voltage is a lot lower. (I think most 2N7002 are likely to work fine.)
I am placing the most critical components first, less critical parts next, and will probably move most of the static logic resistors to the back side of the board to save space.
SMD resistors and caps are 0805, except for some b case tantalums in the digital supplies. I generally prefer 1206 for home projects because they are easier to work with. I find anything smaller than 0603 to be just about impossible to deal with.
I am probably going to replace some of the jumper options with 0805 footprints and depop the parts not needed for the selected option.
I am going to use oscons or BG for the audio bypass and L/R common caps which I have split to in theory reduce LF cross talk between the commons.
Here is the latest rev of the schematic, some additional changes are to make sure that device pin numbers match with the physical shapes used in the pcb.
Still wondering if anyone knows answers to the following questions:
CS8416 - phase detector update rate, normal and fast, what choice did you make and why?
PCM1798 - filter slope. Steep which seems to give the best measured performance vs. the "slow" rolloff which seems to roll off early and exhibits higher thd and noise, but might sound better? Why would this be included otherwise?
Any thoughts are welcome.
Unfortunately that means that some things don't seem to fit too well.
I have eliminated the rectifier, cm choke and large filter cap to save space. I will probably use a cm choke in the raw supply which will be located elsewhere.
I will place all of the voltage regulators which are now all sot-223 sized parts on the back side of the board along with the supply caps.
I have eliminated the external I2S connector as I will never use it, and at higher data rates it represents an unterminated stub.
Changed to the NDS7002 fet (sot-23 package) for the mute led as the threshold voltage is a lot lower. (I think most 2N7002 are likely to work fine.)
I am placing the most critical components first, less critical parts next, and will probably move most of the static logic resistors to the back side of the board to save space.
SMD resistors and caps are 0805, except for some b case tantalums in the digital supplies. I generally prefer 1206 for home projects because they are easier to work with. I find anything smaller than 0603 to be just about impossible to deal with.
I am probably going to replace some of the jumper options with 0805 footprints and depop the parts not needed for the selected option.
I am going to use oscons or BG for the audio bypass and L/R common caps which I have split to in theory reduce LF cross talk between the commons.
Here is the latest rev of the schematic, some additional changes are to make sure that device pin numbers match with the physical shapes used in the pcb.
Still wondering if anyone knows answers to the following questions:
CS8416 - phase detector update rate, normal and fast, what choice did you make and why?
PCM1798 - filter slope. Steep which seems to give the best measured performance vs. the "slow" rolloff which seems to roll off early and exhibits higher thd and noise, but might sound better? Why would this be included otherwise?
Any thoughts are welcome.
kevinkr said:
I chose to pull down TX (phase detector). I'm not sure what "higher phase rate" refers to in the data sheet, so I went with the "normal" option, assuming that SPDIF input from a standard source would be "normal". Someone better informed chime in please!
I'm going with the slow rollof filter, because I have upsampling to 96 kHz. I've just heard so much bad stuff about the "brickwall" filtering, that I figured it was better to minimize those effects, if possible.
I would definitely recommend going with 0805 for all decoupling caps on the IC's. From what I understand, the lower lead inductance on smaller smt caps is a good thing for decoupling purposes.
As for jumpers, I think you should design the board for what you want, not what anyone else could possibly want, unless they are paying for it, of course. Jumpers take up space, and if you want to make the board fit a certain size (as I did), they'd be the first thing I'd remove.
I did finally figure out what the difference is between the two phase detector rates. The two rates have differing spectral distribution of the jitter component, and the fast update rate is recommended for use with sigma delta converters and the normal rate for pcm converters. The point is mute however if sampling rates higher than 96kHz are used as the PLL is not guaranteed to lock in fast update mode above this data rate.
Based on the above and my desire to use this with up to 192kHz sample rates I have eliminated the option to choose one or the other. It's now set for normal updates.
I have configured the CS8416 to handle (the chip will autodetect and switch when required) de-emphasis at 44.1kHz fs automatically, although this is defeatable. The 50uS de-emphasis option is rarely used, but I am sure I have a few cd's that use it, so as a legacy thing I will leave this enabled.
128Fs/256Fs on master clock is selectable, I'm not sure which I will end up using, but both are supported by the 1798 and either will allow operation to 192kHz. I suspect I will probably use 128Fs in order to reduce the maximum clock speed to < 25MHz. (EMI and who knows what other issues lurk in this homebrew pcb.)
The filter is still configurable, but I suspect the faster slope will end up being the choice I use as the response and linearity cited for the slower slope look questionable for use below 96kHz Fs.
This thing runs in hardware mode to obviate the need for a micro controller.
I did manage to purchase a couple of soic version CS8416 from Newark online. The price was reasonable at $6.28 each in single quantities. I wanted to avoid the tssop because of its small size, but it is readily available from Digikey amongst others. I expect to have them early next week.
I already have the PCM1798 (ssop - so small waah) and decided not to pursue the higher performing and more expensive 1794 as I am not using any clock regeneration scheme to reduce mclk jitter and so I suspect the benefits of the higher quality dac are unlikely to be audible.
I/V conversion will either be handled by transformers (Sowter or Llundahl) or by resistor I/V conversion and tube amplification to bring it up to line level.
The board layout is basically complete, and I am going to get some made in a few more weeks.
Here's the latest schematic
Based on the above and my desire to use this with up to 192kHz sample rates I have eliminated the option to choose one or the other. It's now set for normal updates.
I have configured the CS8416 to handle (the chip will autodetect and switch when required) de-emphasis at 44.1kHz fs automatically, although this is defeatable. The 50uS de-emphasis option is rarely used, but I am sure I have a few cd's that use it, so as a legacy thing I will leave this enabled.
128Fs/256Fs on master clock is selectable, I'm not sure which I will end up using, but both are supported by the 1798 and either will allow operation to 192kHz. I suspect I will probably use 128Fs in order to reduce the maximum clock speed to < 25MHz. (EMI and who knows what other issues lurk in this homebrew pcb.)
The filter is still configurable, but I suspect the faster slope will end up being the choice I use as the response and linearity cited for the slower slope look questionable for use below 96kHz Fs.
This thing runs in hardware mode to obviate the need for a micro controller.
I did manage to purchase a couple of soic version CS8416 from Newark online. The price was reasonable at $6.28 each in single quantities. I wanted to avoid the tssop because of its small size, but it is readily available from Digikey amongst others. I expect to have them early next week.
I already have the PCM1798 (ssop - so small waah) and decided not to pursue the higher performing and more expensive 1794 as I am not using any clock regeneration scheme to reduce mclk jitter and so I suspect the benefits of the higher quality dac are unlikely to be audible.
I/V conversion will either be handled by transformers (Sowter or Llundahl) or by resistor I/V conversion and tube amplification to bring it up to line level.
The board layout is basically complete, and I am going to get some made in a few more weeks.
Here's the latest schematic
Definitely missed that one, but yeah that could be a significant benefit.
Still the 4mA pp of the 1798 is significantly better than the 2mA of dacs I have previously converted to passive I/V operation.
Getting married next week-end, and then off on our much deserved honeymoon for a week. I'll be ordering that pcb as soon as I get back.
It's just an experiment to see what I can do with a very simple implementation, and to get my feet wet.
Future ambitions include something with a cypress i2c u controller programmable over usb, separate digital filters, os/nos, parallel dacs, and smart clock regenerator using a vcxo. On reflection that might be too ambitious, I'll take it in baby steps.
Still the 4mA pp of the 1798 is significantly better than the 2mA of dacs I have previously converted to passive I/V operation.
Getting married next week-end, and then off on our much deserved honeymoon for a week. I'll be ordering that pcb as soon as I get back.
It's just an experiment to see what I can do with a very simple implementation, and to get my feet wet.
Future ambitions include something with a cypress i2c u controller programmable over usb, separate digital filters, os/nos, parallel dacs, and smart clock regenerator using a vcxo. On reflection that might be too ambitious, I'll take it in baby steps.
Hey Steve,
Thanks for the tips, I'll move those resistors, value isn't cast in stone either. I was thinking back termination, but then stuck them in the wrong spot. Most apps I've seen recommended 47 - 51 ohms, any particular reason you recommend the 22's? (I haven't attempted to make the traces a 50 ohm transmission line over the short distances required.)
The cap on the aes ground was deliberate and I know not common practice - I was attempting to avoid interconnection of device grounds between source and the dac. If this should cause a problem I will replace the cap with either a small resistor or a zero ohm jumper.
Thanks for the tips, I'll move those resistors, value isn't cast in stone either. I was thinking back termination, but then stuck them in the wrong spot. Most apps I've seen recommended 47 - 51 ohms, any particular reason you recommend the 22's? (I haven't attempted to make the traces a 50 ohm transmission line over the short distances required.)
The cap on the aes ground was deliberate and I know not common practice - I was attempting to avoid interconnection of device grounds between source and the dac. If this should cause a problem I will replace the cap with either a small resistor or a zero ohm jumper.
Hi all,
earlier in this thread, the "steep vs. slow filter question" for the PCM1794 was raised. I'd like to know if anybody out there knows more about the different filters.
When I connect my Philips CDPro2M to a PCM1794a via I2S , I observed that the square wave response looked really strange when selecting the "slow" filter.
Imagine square wave responses from about 10 different filter types (Chebycheff, Bessel, Butterworth, etc...) and put them on top of each other. It ends up looking like a signal that can choose from a limited number of different ways to approximate the square wave and when looking to the scope, they form a regular pattern. I don't have a digital scope so I can' t analyze further.
Anyway, it looked ok with "steep" filter. It also looked ok with "slow" filter and the drive's digital filter doing 2x oversampling. Does anybody know what causes that strange pattern in the "slow" filter's square wave response?
Kind regards
Sebastian
earlier in this thread, the "steep vs. slow filter question" for the PCM1794 was raised. I'd like to know if anybody out there knows more about the different filters.
When I connect my Philips CDPro2M to a PCM1794a via I2S , I observed that the square wave response looked really strange when selecting the "slow" filter.
Imagine square wave responses from about 10 different filter types (Chebycheff, Bessel, Butterworth, etc...) and put them on top of each other. It ends up looking like a signal that can choose from a limited number of different ways to approximate the square wave and when looking to the scope, they form a regular pattern. I don't have a digital scope so I can' t analyze further.
Anyway, it looked ok with "steep" filter. It also looked ok with "slow" filter and the drive's digital filter doing 2x oversampling. Does anybody know what causes that strange pattern in the "slow" filter's square wave response?
Kind regards
Sebastian
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.