Hi soundcheck,
I am no longer using any computer output (SPDIF nor USB) as I can't get bit-perfect output from my iMac / Mac OSX anymore, after recent software updates (iTunes / Quicktime), the soundmixer now seems to be active all the time (up/down sampling), and prevents bit-perfect playback, this causes significant sound quality degradation (very clearly audible). I suspect usable output resolution is now reduced to 14 .. 15 bits.
It seems that achieving bit-perfect playback requires constant tweaking / monitoring, and I just don't like that.
So I am now using Apple airport express module. It does provide bit-perfect output regardless of OS / application, as it just receives files over the (wireless) network.
I currently use a CS8412 in slave clock mode (both BCK and WS are now inputs). MCLK is still output, and outputs 256 * fs (source clock). So when divided by 4, this signal (2.8224 MHz) can be used as input timing reference while the CS8412 runs in slave clock mode.
I designed and built a discrete very low power VCXO running @ 11.2896 MHz, with approx. +/- 100ppm control range. The output is divided by 4 in order to get 2.8224 MHz. Multiple clock buffers are connected directly to this divider output. The 4 DAC chips receive clean 2.8224 MHz symmetrical square wave clocks through dedicated UHS clock buffers.
I used short miniature coax interlinks for transporting BCK from clock buffer directly to the DAC BCK input, and applied correct impedance matching.
I used the phase discriminator section from a 74HC4046 (32 MHz version), and disabled the VCO. Next I fed both the 2.8224 MHz VCXO clock, and the 2.8224 MHz input reference clock (derived from MCK) into it. The output feeds a loop filter. The loop filter drives the VCXO control input.
So the VCXO frequency is now slowly adjusted until it's in perfect sync / phase with the SPDIF source clock. During lock-in (few seconds), there is still sound as the CS8412 runs in slave clock mode, but when playing-back a constant frequency during this locked-out condition, periodic clicks could be heard..
Different sample rates can be supported by using appropriate VCXO crystal frequency.
I am no longer using any computer output (SPDIF nor USB) as I can't get bit-perfect output from my iMac / Mac OSX anymore, after recent software updates (iTunes / Quicktime), the soundmixer now seems to be active all the time (up/down sampling), and prevents bit-perfect playback, this causes significant sound quality degradation (very clearly audible). I suspect usable output resolution is now reduced to 14 .. 15 bits.
It seems that achieving bit-perfect playback requires constant tweaking / monitoring, and I just don't like that.
So I am now using Apple airport express module. It does provide bit-perfect output regardless of OS / application, as it just receives files over the (wireless) network.
I currently use a CS8412 in slave clock mode (both BCK and WS are now inputs). MCLK is still output, and outputs 256 * fs (source clock). So when divided by 4, this signal (2.8224 MHz) can be used as input timing reference while the CS8412 runs in slave clock mode.
I designed and built a discrete very low power VCXO running @ 11.2896 MHz, with approx. +/- 100ppm control range. The output is divided by 4 in order to get 2.8224 MHz. Multiple clock buffers are connected directly to this divider output. The 4 DAC chips receive clean 2.8224 MHz symmetrical square wave clocks through dedicated UHS clock buffers.
I used short miniature coax interlinks for transporting BCK from clock buffer directly to the DAC BCK input, and applied correct impedance matching.
I used the phase discriminator section from a 74HC4046 (32 MHz version), and disabled the VCO. Next I fed both the 2.8224 MHz VCXO clock, and the 2.8224 MHz input reference clock (derived from MCK) into it. The output feeds a loop filter. The loop filter drives the VCXO control input.
So the VCXO frequency is now slowly adjusted until it's in perfect sync / phase with the SPDIF source clock. During lock-in (few seconds), there is still sound as the CS8412 runs in slave clock mode, but when playing-back a constant frequency during this locked-out condition, periodic clicks could be heard..
Different sample rates can be supported by using appropriate VCXO crystal frequency.
Hi tubee,
As long as you are just moving files, the OS will make sure data integrity is maintained.
The problematic part is getting this exact digital audio data into the DAC without software updates, OS, applications, drivers, settings, digital audio transmitter / digital audio receiver messing it up.
Best would be to just bypass all this mess, and leave digital audio file > digital audio stream conversion to dedicated hardware.
As long as you are just moving files, the OS will make sure data integrity is maintained.
The problematic part is getting this exact digital audio data into the DAC without software updates, OS, applications, drivers, settings, digital audio transmitter / digital audio receiver messing it up.
Best would be to just bypass all this mess, and leave digital audio file > digital audio stream conversion to dedicated hardware.
PC as source and other various
that is an interesting theory - but for me in practice my PC provides a very good quality data stream via the SPDIF which combined with a single 1543 DAC gives clarity & transparency that I have rarely heard.
I am still not happy with the sound though, too much distortion.
Also I compared foobar with XX highend on windows XP media edition and I could hear absolutely no difference so if anyone can tell me what I have to do to get xx highend sounding better I would be interested to hear about it.
My biggest improvement came from keeping the digital leads VERY short
I am using caddock TF020's ( think that's right ) in a few critical places and they are excellent !
cheers
mike
that is an interesting theory - but for me in practice my PC provides a very good quality data stream via the SPDIF which combined with a single 1543 DAC gives clarity & transparency that I have rarely heard.
I am still not happy with the sound though, too much distortion.
Also I compared foobar with XX highend on windows XP media edition and I could hear absolutely no difference so if anyone can tell me what I have to do to get xx highend sounding better I would be interested to hear about it.
My biggest improvement came from keeping the digital leads VERY short
I am using caddock TF020's ( think that's right ) in a few critical places and they are excellent !
cheers
mike
Dear -ecdesigns-,
While I admire your intellectual restlessness
I must declare that I am in mood different from the digital tweaker, now.
In fact I am so delighted by the music I'm getting through my modded DI16 that I only want to build non-digital stuff ...some amps, my last horn speakers and some DIY interconnects (same as yours but with OCC (mono-crystal ) copper, of which I must be the only fan), this year...
(BTW, did you gave up with your research about current sources to tame thermal memory on silicon chips???)
Having said that, and as an exercise, how would this configuration work?
DI16: two groups of 8 TDA1543 DACs; 4.6V or 5V supplies; passive I/V; 500mV balanced output--> summing and amplifying opamp-> Rout, without attenuation.
I/V R=> 0.5V/8*0.0022= 28.4R Moebious loop type resistor, ideally OCC copper , if I ever realize how ***** it is constructed 
(same type of calculation for 8*3 or 8*4 DACs)
The tough part: (at least for me)
If I understood well, Ref pin not connected to ground through Rref but connected to a Voltage reference like LM336 (+2.5V)?. One LM336 per DAC group? (10mA max)
Good! Less parts, less money involved. 
Cheers,
M
While I admire your intellectual restlessness
I must declare that I am in mood different from the digital tweaker, now. In fact I am so delighted by the music I'm getting through my modded DI16 that I only want to build non-digital stuff
...some amps, my last horn speakers and some DIY interconnects (same as yours but with OCC (mono-crystal ) copper, of which I must be the only fan), this year...(BTW, did you gave up with your research about current sources to tame thermal memory on silicon chips???)
Having said that, and as an exercise, how would this configuration work?
DI16: two groups of 8 TDA1543 DACs; 4.6V or 5V supplies; passive I/V; 500mV balanced output--> summing and amplifying opamp-> Rout, without attenuation.
I/V R=> 0.5V/8*0.0022= 28.4R Moebious loop type resistor, ideally OCC copper
, if I ever realize how ***** it is constructed (same type of calculation for 8*3 or 8*4 DACs)
The tough part: (at least for me)
If I understood well, Ref pin not connected to ground through Rref but connected to a Voltage reference like LM336 (+2.5V)?. One LM336 per DAC group? (10mA max)
Good! Less parts, less money involved.
Cheers,
M
tubee said:
Kind of OFF-TOPIC
:We had this discussion some weeks ago in the Linux-Audio thread:
Install Wine under Linux ($sudo apt-get install wine).
Run wine-config ($winecfg) and setup your drives
Then you download EAC and just run the installation
under Linux.
The first thing you need to do. Skip the configuration wizard
and switch to win32 instead of ASPI under Interfaces.
Exit EAC and restart. Now you can configure it.
It works really well.
Since the Accurate Rip feature is working, you can bet that 90% of your rips are OK.
Soon you'll see the instructions in the Linux-Audio Wiki.
@ec-designs:
I still believe that it's not too difficult to get bit-perfect data (WHATEVER IT MEANS - BER<1*10-12 !?!?
) from Linux. Good luck.
Cheers
Thanks Ecdesigns and Souncheck for reply's
Maybe off topic, but here not really
I am considering a good dac when i have some money, or revive my creatures (PCM56 passive IV & tube output) l consider a new silent linux PC here in the listening room, and i am very confident linux is THE system to work with.
Adding a separate SPDif soundcard is also a possibility, and then reclock the signal after CS8414 receiverchip. I wonder if a cheap sweex usb soundcard with SPDif output sounds properly. (after reclocking!)
I have Wine tried allready on the upstairs PC, unfortunately sketchup doesn't work proper.
Maybe off topic, but here not really
I am considering a good dac when i have some money, or revive my creatures (PCM56 passive IV & tube output) l consider a new silent linux PC here in the listening room, and i am very confident linux is THE system to work with.
Adding a separate SPDif soundcard is also a possibility, and then reclock the signal after CS8414 receiverchip. I wonder if a cheap sweex usb soundcard with SPDif output sounds properly. (after reclocking!)
I have Wine tried allready on the upstairs PC, unfortunately sketchup doesn't work proper.
Hi soundcheck,
With bit-perfect playback I simply mean that I get the exact same sample data at the USB / SPDIF output as contained in the original digital audio file. As I noted before, this looks easier than it actually is.
If digital audio data is NOT bit perfect, it usually means that ALL sample values have (slightly) changed, just like with digital brickwall filtering.
This might not seem to be such a big issue, until you hear the effect on sound quality.
With bit-perfect playback I simply mean that I get the exact same sample data at the USB / SPDIF output as contained in the original digital audio file. As I noted before, this looks easier than it actually is.
If digital audio data is NOT bit perfect, it usually means that ALL sample values have (slightly) changed, just like with digital brickwall filtering.
This might not seem to be such a big issue, until you hear the effect on sound quality.
Hi mikelm,
This depends how SPDIF is derived from the PC, if a dedicated sound-card, and proper settings / drivers are used, output may already be bit-perfect.
I use the TOSLINK output from a CD reference transport for comparison. The reason I switched to TOSLINK is because I want to have full galvanic insulation, and zero coupling capacitance between digital audio transmitter and DAC. When using SPDIF coax, it would be virtually impossible to achieve this, creating a low impedance path for HF interference.
When using TOSLINK it's important to block source clock jitter effectively.
Since DAC THD is mainly caused by bit errors, and you are using a single TDA1543 in NOS mode, it will have an error of roughly (THD / 100) * bit resolution = (0.018 / 100) * 2^16 = 6.5, bit error would then be approx. 2.75. So this configuration provides 16 - 2.75 = 13.25 bit resolution typical. This resolution makes it very difficult to tell the difference between bit-perfect playback (16 bits) and a manipulated digital audio data (with typical usable bit resolution of 12 ... 15).
Similar TDA1541A error would be roughly (0.0018 / 100) * 2^16 = 1.17, bit error would be 1. so it would provide 16-1 = 15 bit accuracy typical. The DI4T uses 4 x TDA1541A, this reduces THD (bit errors) by SQRT(4) = 2. So theoretical THD would now be 0.0018 / 2 = 0.0009%. (0.0009 / 100) * 2^16 = 0.58, approx. 0.5 bit error, or a usable bit resolution of 15.5 bits. Since I use 4 x direct interpolation, and each TDA1541A provides 15 bit typical accuracy, effective bit resolution would be approx. 17 bits.
Now the difference between a bit-perfect (16 bits) and manipulated data (14 ... 15 bits) becomes VERY clearly audible.
If XXHighend manages to provide bit-perfect playback (through USB), using the correct settings, and you have an audio set capable of resolving 15 bits or more (your current set only resolves 13.25 bits typical, due to the single TDA1543), the difference in sound quality should be clearly audible.
This depends how SPDIF is derived from the PC, if a dedicated sound-card, and proper settings / drivers are used, output may already be bit-perfect.
I use the TOSLINK output from a CD reference transport for comparison. The reason I switched to TOSLINK is because I want to have full galvanic insulation, and zero coupling capacitance between digital audio transmitter and DAC. When using SPDIF coax, it would be virtually impossible to achieve this, creating a low impedance path for HF interference.
When using TOSLINK it's important to block source clock jitter effectively.
Since DAC THD is mainly caused by bit errors, and you are using a single TDA1543 in NOS mode, it will have an error of roughly (THD / 100) * bit resolution = (0.018 / 100) * 2^16 = 6.5, bit error would then be approx. 2.75. So this configuration provides 16 - 2.75 = 13.25 bit resolution typical. This resolution makes it very difficult to tell the difference between bit-perfect playback (16 bits) and a manipulated digital audio data (with typical usable bit resolution of 12 ... 15).
Similar TDA1541A error would be roughly (0.0018 / 100) * 2^16 = 1.17, bit error would be 1. so it would provide 16-1 = 15 bit accuracy typical. The DI4T uses 4 x TDA1541A, this reduces THD (bit errors) by SQRT(4) = 2. So theoretical THD would now be 0.0018 / 2 = 0.0009%. (0.0009 / 100) * 2^16 = 0.58, approx. 0.5 bit error, or a usable bit resolution of 15.5 bits. Since I use 4 x direct interpolation, and each TDA1541A provides 15 bit typical accuracy, effective bit resolution would be approx. 17 bits.
Now the difference between a bit-perfect (16 bits) and manipulated data (14 ... 15 bits) becomes VERY clearly audible.
If XXHighend manages to provide bit-perfect playback (through USB), using the correct settings, and you have an audio set capable of resolving 15 bits or more (your current set only resolves 13.25 bits typical, due to the single TDA1543), the difference in sound quality should be clearly audible.
Hi maxlorenz,
My aim is to design and build the ultimate (NOS) DAC, this proves to be an extremely difficult task, but I am determined to achieve it.
Current sources can only pull the OP-amp output stage in class A (higher bias current), this would theoretically reduce crossover distortion only.
The most crucial part of any (OP) amp is the differential input stage, and I can't tweak that.
Audio OP-amps have so many issues, that it's better to use discrete circuits for High-End audio applications instead. The major advantage of using discrete JFET or tube amplifiers is increased resolution. Note that I am no longer aiming for vanishing low THD, as this has nothing to do with resolution nor dynamic performance.
It might work better than expected, but it's best to use 4 x interpolation (4 x TDA1543 placed in the upper row).
I/V resistor value : 4 x 110R (leave TDA1543 outputs in parallel).
Diff. stage resistor values: 1K96 / 22K
Diff. stage capacitors : 100pF (use from Op-amp I/V).
This should provide approx. 2V rms output amplitude. The attenuator can now be left out.
For best performance, discrete JFET (borbely audio) or tube differential amplifiers are required.
It's best to use only 4 selected TDA1543 chips that have lowest bit errors. I had good results with TDA1543 chips made in Taiwan.
Not quite, the external 2.5V reference voltage is used for passive I/V resistor reference (you can't connect the passive I/V resistors to GND with this configuration). All TDA1543 Vref pins should remain unconnected (remove 4K7 resistors). Since OP-amp I/V converters are no longer used, TDA1543 Vref is no longer required.
My aim is to design and build the ultimate (NOS) DAC, this proves to be an extremely difficult task, but I am determined to achieve it.
Current sources can only pull the OP-amp output stage in class A (higher bias current), this would theoretically reduce crossover distortion only.
The most crucial part of any (OP) amp is the differential input stage, and I can't tweak that.
Audio OP-amps have so many issues, that it's better to use discrete circuits for High-End audio applications instead. The major advantage of using discrete JFET or tube amplifiers is increased resolution. Note that I am no longer aiming for vanishing low THD, as this has nothing to do with resolution nor dynamic performance.
It might work better than expected, but it's best to use 4 x interpolation (4 x TDA1543 placed in the upper row).
I/V resistor value : 4 x 110R (leave TDA1543 outputs in parallel).
Diff. stage resistor values: 1K96 / 22K
Diff. stage capacitors : 100pF (use from Op-amp I/V).
This should provide approx. 2V rms output amplitude. The attenuator can now be left out.
For best performance, discrete JFET (borbely audio) or tube differential amplifiers are required.
It's best to use only 4 selected TDA1543 chips that have lowest bit errors. I had good results with TDA1543 chips made in Taiwan.
Not quite, the external 2.5V reference voltage is used for passive I/V resistor reference (you can't connect the passive I/V resistors to GND with this configuration). All TDA1543 Vref pins should remain unconnected (remove 4K7 resistors). Since OP-amp I/V converters are no longer used, TDA1543 Vref is no longer required.
Hi EC,
I'm a bit confused...again. You said you tried the DI8 with passive I/V + tube output, but it doesn't sound as good as your DI4T. I'm assuming the only difference in these two design is the number of DAC chips. Can you explain? Did you used the DI8 PCB for DI4 or do you have a new PCB? Thanks.
I'm a bit confused...again. You said you tried the DI8 with passive I/V + tube output, but it doesn't sound as good as your DI4T. I'm assuming the only difference in these two design is the number of DAC chips. Can you explain? Did you used the DI8 PCB for DI4 or do you have a new PCB? Thanks.
Thanks -ecdesigns- for your reply. (Post #1970)
I know, I know...and I (we) am (are) indebted for it...
I suppose this will give better HF extension.
But instead of 4 chips in a row, isn't it better to put 4 "towers" of 4 (or 8) chips each, two in each row (row means "DAC group", right?). With corresponding I/V R reduction.
Thanks for the current reference and Rref explanation.
Will one LM336 be enough for all 4 I/V R?
Many thanks.
M
I know, I know...and I (we) am (are) indebted for it...
I suppose this will give better HF extension.
But instead of 4 chips in a row, isn't it better to put 4 "towers" of 4 (or 8) chips each, two in each row (row means "DAC group", right?). With corresponding I/V R reduction.
Thanks for the current reference and Rref explanation.
Will one LM336 be enough for all 4 I/V R?
Many thanks.
M
DI16 power supply schematics
Hi luvdunhill,
Use the value indicated in the documentation (2 x 18V), the picture is only for illustration.
I attached DI16 power supply schematics.
Both E1 & E2 are used for mains voltage selection. The thick line on the schematics indicates 230V setting (both primary windings in series). For 115V, leave-out the connection indicated by the thick line, and use the connections indicated by the two dashed lines. This puts both primary windings in parallel.
Hi luvdunhill,
Use the value indicated in the documentation (2 x 18V), the picture is only for illustration.
I attached DI16 power supply schematics.
Both E1 & E2 are used for mains voltage selection. The thick line on the schematics indicates 230V setting (both primary windings in series). For 115V, leave-out the connection indicated by the thick line, and use the connections indicated by the two dashed lines. This puts both primary windings in parallel.
Attachments
-ecdesigns- said:
Yes, I'm finding the same thing:
http://www.diyaudio.com/forums/showthread.php?s=&postid=1480424#post1480424
-ecdesigns- said:
But this limits the bit depth and sample rate, no? With the arrival of hires downloads, this is a shame.
Does a simple 2-channel soundcard with spdif output, wordclock input and no soundmixer exist? How difficult would it be to build?
Finally, ecdesigns, do your commercial nos dacs have wordclock outputs?
Mani.
Hi MGH,
No, there are much more differences:
- SPDIF (TOSLINK) input only
- VCXO 11.2896 MHz masterclock, CS8412/16 in slave clock mode
- Dedicated clock buffers for DAC chips, CS8412/16, timing module and phase discriminator
- All TDA1541A chips are now clocked simultaneously (no BCK inverter on DAC1)
- 4 x TDA1541A (4 x interpolation)
- Exlusive use of WIMA 1uF miniature foil caps for decoupling (DA1541A modules)
- I2S attenuators / level shifters on each DAC chip I2S input
- Differential passive I/V conversion
- Tube-only output, (increased gain, higher twin cathode follower bias, gyrators & current sources removed)
- Bulk metal foil resistors in tube diff. input stage
- output attenuators removed (output impedance approx. 220R)
- Modified HV Teddyreg in the tube power supply
4 x interpolation reduces jitter sensitivity, based on 15 bit effective resolution for each TDA1541A chip (1 LSB error typical). Jitter sensitivity is now 1/(176,400*2^17)/2 = 21.6ps. 8 x interpolation jitter sensitivity equals 1/(352,800*2^18)/2 = 5.4ps
The reduced linear interpolation factor reduces distortion introduced by linear interpolation, and changes HF rolloff (interpolation filter characteristics).
Fewer DAC chips also result in less crosstalk.
Other side effects are reduced power dissipation, and reduce costs (fewer modules required).
I built my DI4T prototype using existing DI8M PCBs.
No, there are much more differences:
- SPDIF (TOSLINK) input only
- VCXO 11.2896 MHz masterclock, CS8412/16 in slave clock mode
- Dedicated clock buffers for DAC chips, CS8412/16, timing module and phase discriminator
- All TDA1541A chips are now clocked simultaneously (no BCK inverter on DAC1)
- 4 x TDA1541A (4 x interpolation)
- Exlusive use of WIMA 1uF miniature foil caps for decoupling (DA1541A modules)
- I2S attenuators / level shifters on each DAC chip I2S input
- Differential passive I/V conversion
- Tube-only output, (increased gain, higher twin cathode follower bias, gyrators & current sources removed)
- Bulk metal foil resistors in tube diff. input stage
- output attenuators removed (output impedance approx. 220R)
- Modified HV Teddyreg in the tube power supply
4 x interpolation reduces jitter sensitivity, based on 15 bit effective resolution for each TDA1541A chip (1 LSB error typical). Jitter sensitivity is now 1/(176,400*2^17)/2 = 21.6ps. 8 x interpolation jitter sensitivity equals 1/(352,800*2^18)/2 = 5.4ps
The reduced linear interpolation factor reduces distortion introduced by linear interpolation, and changes HF rolloff (interpolation filter characteristics).
Fewer DAC chips also result in less crosstalk.
Other side effects are reduced power dissipation, and reduce costs (fewer modules required).
I built my DI4T prototype using existing DI8M PCBs.
Dear -ecdesings-,
Could you please post a schematic of the I2S attenuators/level shifters...I read your previous post about this topic but I fail to see how they are connected.
Also, what does this "simultaneous clocking" mean?
Compared to the previous situation, that is...
Cheers,
M
PS: your capability of re-studying and re-formulating the whole approach and also your willingness to listen and consider critics/advices, apart being very commendable, has narrowed my options about trying to guess what your astrological sign is
This is a hobby of mine and I am very bad a it...
Could you please post a schematic of the I2S attenuators/level shifters...I read your previous post about this topic but I fail to see how they are connected.
Also, what does this "simultaneous clocking" mean?
Compared to the previous situation, that is...
Cheers,
M
PS: your capability of re-studying and re-formulating the whole approach and also your willingness to listen and consider critics/advices, apart being very commendable, has narrowed my options about trying to guess what your astrological sign is
This is a hobby of mine and I am very bad a it...
Did someone try Logitech Squeezebox Duet
it seems to be an interesting alternative to Airport Express.
For Airport Express a Computer in the room is needed to control iTunes. Squeezebox comes with a wireless controller.
This results in some advantages:
- only one computer needed for more rooms
- no need for a super silent pc
- a large number of OS supported
Does someone have experience with the new Squeezebox?
Ciao
Peter
it seems to be an interesting alternative to Airport Express.
For Airport Express a Computer in the room is needed to control iTunes. Squeezebox comes with a wireless controller.
This results in some advantages:
- only one computer needed for more rooms
- no need for a super silent pc
- a large number of OS supported
Does someone have experience with the new Squeezebox?
Ciao
Peter