I was not aware of the Odroid-HC1.
ODROID | Hardkernel
This makes a powerful and inexpensive NAS. There is a HC2 coming in November for 3.5' drives.
I just ordered one and a 2TB HDD and I will report how it goes.
ODROID | Hardkernel
This makes a powerful and inexpensive NAS. There is a HC2 coming in November for 3.5' drives.
I just ordered one and a 2TB HDD and I will report how it goes.
Thanx a lot. I´ll definitely have a look at this. Should run Minimserver smoothly...
Cheers
Hi bernie_frank,
The Host (computer) produces a noise spectrum, this spectrum also appears on the USB bus voltage and USB serial data lines. This spectrum mixes with the spectrum produced by the data being sent over the USB interface. The USB interlink mixes USB bus noise and USB data noise and applies filtering (low pass) due to stray capacitance and inductance.
OS and all the applications that run on it will produce a unique noise spectrum. When the OS and or applications are changed, this noise spectrum changes with it. This can be verified by connecting a suitable spectrum analyser to the USB bus voltage and USB serial data lines. When changes are made in the OS and or application we can measure that the noise spectrum changes.
The problem is that we cannot block this noise spectrum in a DAC because the digital audio interface that feeds data into the DAC has a non zero bandwidth of up to several hundred MHz. This allows the noise spectrum to enter the DAC together with the audio data. If we block the noise spectrum we won't be able to pass the data. If we limit the bandwidth we risk data corruption.
The noise spectrum finds its way to the masterclock, ground plane, power supplies and finally the D/A converter output. The combination of this host noise plus the noise generated and used by the DAC circuits leads to a change in the DAC output spectrum. Now we no longer hear only what's on the recording but also the interference that folds back into the audio spectrum.
This added distortion varies with the music content (related data pattern), so some distortion only becomes audible with specific music.
By reducing and shaping the host noise spectrum we can improve the sound quality produced by the DAC.
Depending on the audio set we might create a ground loop that feeds even more noise into the DAC and into the connected audio equipment. This problem is usually fixed by using USB isolators or isolators further up in the digital signal path.
These isolators always produce jitter (100ps and up) and add noise. This jitter and noise mixes with the already present interference spectra, causing more degrading. Even if by some miracle, sample timing is not changed by this USB jitter, the interference that spreads over the other DAC circuits will still cause degrading.
Isolators won't block host noise as these isolators need to offer sufficient bandwidth for the data to pass. This way interference can still pass together with the audio data.
USB isolators have certain stray capacitance between both input and output. This enables very high frequencies to simply bypass the isolator. Ferrite beads can be used to attenuate these very high frequencies.
Another option to avoid ground loops:
WIFI (digital audio data) -> battery powered laptop -> USB -> DAC -> Power amplifiers -> speakers.
The Host (computer) produces a noise spectrum, this spectrum also appears on the USB bus voltage and USB serial data lines. This spectrum mixes with the spectrum produced by the data being sent over the USB interface. The USB interlink mixes USB bus noise and USB data noise and applies filtering (low pass) due to stray capacitance and inductance.
OS and all the applications that run on it will produce a unique noise spectrum. When the OS and or applications are changed, this noise spectrum changes with it. This can be verified by connecting a suitable spectrum analyser to the USB bus voltage and USB serial data lines. When changes are made in the OS and or application we can measure that the noise spectrum changes.
The problem is that we cannot block this noise spectrum in a DAC because the digital audio interface that feeds data into the DAC has a non zero bandwidth of up to several hundred MHz. This allows the noise spectrum to enter the DAC together with the audio data. If we block the noise spectrum we won't be able to pass the data. If we limit the bandwidth we risk data corruption.
The noise spectrum finds its way to the masterclock, ground plane, power supplies and finally the D/A converter output. The combination of this host noise plus the noise generated and used by the DAC circuits leads to a change in the DAC output spectrum. Now we no longer hear only what's on the recording but also the interference that folds back into the audio spectrum.
This added distortion varies with the music content (related data pattern), so some distortion only becomes audible with specific music.
By reducing and shaping the host noise spectrum we can improve the sound quality produced by the DAC.
Depending on the audio set we might create a ground loop that feeds even more noise into the DAC and into the connected audio equipment. This problem is usually fixed by using USB isolators or isolators further up in the digital signal path.
These isolators always produce jitter (100ps and up) and add noise. This jitter and noise mixes with the already present interference spectra, causing more degrading. Even if by some miracle, sample timing is not changed by this USB jitter, the interference that spreads over the other DAC circuits will still cause degrading.
Isolators won't block host noise as these isolators need to offer sufficient bandwidth for the data to pass. This way interference can still pass together with the audio data.
USB isolators have certain stray capacitance between both input and output. This enables very high frequencies to simply bypass the isolator. Ferrite beads can be used to attenuate these very high frequencies.
Another option to avoid ground loops:
WIFI (digital audio data) -> battery powered laptop -> USB -> DAC -> Power amplifiers -> speakers.
Hi Venusfly,
These are active divider decoupling caps. Active dividers are part of the DEM circuit used in the TDA1541A.
DEM stand for Dynamic Element matching. The idea is to take say 4 unequal currents (component tolerances) and create 4 identical currents from these.
Suppose we created 4 current sources in a chip and end up with 1mA, 0.8mA, 1.3mA and 0.9mA. Without further action this would lead to bit errors and distortion.
Some D/A converter manufacturers fixed this problem using laser trimming, but this is rather expensive. Despite laser trimming, component tolerances may change over time requiring external (re) adjustment. Usually there is a MSB trim input available for this.
Philips wanted to lower the price (no laser adjustment) and maintain high accuracy regardless of component ageing and came up with this DEM trick.
We could add all 4 currents together 1 + 0.8 + 1.3 + 0.9 = 4mA. Then divide this 4mA by 4. This way we could generate 4 equal currents, in this case 1mA from 4 input currents that are not equal. This is basically what DEM does, it creates 4 matched output currents from 4 unequal input currents.
What we need are binary weighted currents of say 2mA, 1mA, 0.5mA, 0.25mA, 0.125mA and so on We need 16 binary weighted currents in order to construct a 16 bit D/A converter.
Philips uses dynamic element matching for getting equal currents and current combination and division of these equal currents in order to get highly accurate binary weighted currents. By combining these currents with electronic switches we can generate 16 bit output current with 65,536 different current levels.
Active divider circuits perform addition and division of currents electronically. In order to do so it has to switch between 4 input currents and 4 output currents using a switch matrix. The output of this switch matrix (active divider) produces ripple current. Ripple current increases as tolerances between the input currents increases.
Decoupling caps are used to reduce this ripple current. There are multiple active dividers in the TDA1541A so we need multiple decoupling caps.
The switch matrix is driven by a shift register that provides required timing signals for operating the switch matrix. The shift register is clocked by a free running on-chip oscillator called the DEM clock.
If the DEM circuit fails, bit currents won't be matched and we end up with (low level) distortion. This distortion can be measured using a -60dB test track for example.
The DEM clock typically operates between approx. 200 and 250 KHz (Radio Frequency) and the ripple frequency on the active dividers will have radio frequency too. The DEM clock circuit requires one external timing cap between pins 16 & 17. With the datasheet application there will be very high jitter on this DEM clock when audio data is processed by the TDA1541A. This gives us an impression of the on-chip crosstalk.
If one wants to minimise this jitter one can add one 6K8 resistor between pin 15 and pin 16 and one 6K8 resistor between pin 15 and pin 17. If one is happy with the high deterministic (data related) jitter on the DEM clock then simply use the datasheet application.
In order to filter out this active divider ripple we need tiny, low inductance RF decoupling caps, no bulky audio caps with long wires.
However, decoupling will not eliminate the interference, it simply moves it somewhere else. You suppress the signal on the active divider output pin and it pops up in another location. One possible way to get rid of interference is to turn it into heat. In that case we need lossy materials (ferrite beads) that turn interference into heat.
Similar when you use a voltage regulator that powers a noisy source, the interference will pop up again at the input of the voltage regulator. If we decouple the interference at the input of the voltage regulator, part of this interference pops back up at the voltage regulator output.
So the theoretically correct decoupling may not work that well in practice. Sure we measure lower ripple voltage on the active divider pin with a scope, but where did the ripple go? It simply moved to a different location where it could possibly do even more harm.
There are DIY members reporting improved sound quality when leaving out these decoupling caps altogether.
So what's the worse thing that could happen when we leave out all DEM caps? We end up with some radio frequency ripple on the output signal. This radio frequency can be dissipated by means of suitable lossy ferrite beads and we end up with the averaged current again.
If we increase the DEM clock frequency by using say 100pF (roughly 1 MHz DEM clock) timing cap, we could increase the ripple frequency so it could be dissipated more easily when using ferrite beads.
The ferrite beads could be placed between TDA1541A output and a passive I/V resistor or between the TDA1541A output and active I/V converter circuit.
These are active divider decoupling caps. Active dividers are part of the DEM circuit used in the TDA1541A.
DEM stand for Dynamic Element matching. The idea is to take say 4 unequal currents (component tolerances) and create 4 identical currents from these.
Suppose we created 4 current sources in a chip and end up with 1mA, 0.8mA, 1.3mA and 0.9mA. Without further action this would lead to bit errors and distortion.
Some D/A converter manufacturers fixed this problem using laser trimming, but this is rather expensive. Despite laser trimming, component tolerances may change over time requiring external (re) adjustment. Usually there is a MSB trim input available for this.
Philips wanted to lower the price (no laser adjustment) and maintain high accuracy regardless of component ageing and came up with this DEM trick.
We could add all 4 currents together 1 + 0.8 + 1.3 + 0.9 = 4mA. Then divide this 4mA by 4. This way we could generate 4 equal currents, in this case 1mA from 4 input currents that are not equal. This is basically what DEM does, it creates 4 matched output currents from 4 unequal input currents.
What we need are binary weighted currents of say 2mA, 1mA, 0.5mA, 0.25mA, 0.125mA and so on We need 16 binary weighted currents in order to construct a 16 bit D/A converter.
Philips uses dynamic element matching for getting equal currents and current combination and division of these equal currents in order to get highly accurate binary weighted currents. By combining these currents with electronic switches we can generate 16 bit output current with 65,536 different current levels.
Active divider circuits perform addition and division of currents electronically. In order to do so it has to switch between 4 input currents and 4 output currents using a switch matrix. The output of this switch matrix (active divider) produces ripple current. Ripple current increases as tolerances between the input currents increases.
Decoupling caps are used to reduce this ripple current. There are multiple active dividers in the TDA1541A so we need multiple decoupling caps.
The switch matrix is driven by a shift register that provides required timing signals for operating the switch matrix. The shift register is clocked by a free running on-chip oscillator called the DEM clock.
If the DEM circuit fails, bit currents won't be matched and we end up with (low level) distortion. This distortion can be measured using a -60dB test track for example.
The DEM clock typically operates between approx. 200 and 250 KHz (Radio Frequency) and the ripple frequency on the active dividers will have radio frequency too. The DEM clock circuit requires one external timing cap between pins 16 & 17. With the datasheet application there will be very high jitter on this DEM clock when audio data is processed by the TDA1541A. This gives us an impression of the on-chip crosstalk.
If one wants to minimise this jitter one can add one 6K8 resistor between pin 15 and pin 16 and one 6K8 resistor between pin 15 and pin 17. If one is happy with the high deterministic (data related) jitter on the DEM clock then simply use the datasheet application.
In order to filter out this active divider ripple we need tiny, low inductance RF decoupling caps, no bulky audio caps with long wires.
However, decoupling will not eliminate the interference, it simply moves it somewhere else. You suppress the signal on the active divider output pin and it pops up in another location. One possible way to get rid of interference is to turn it into heat. In that case we need lossy materials (ferrite beads) that turn interference into heat.
Similar when you use a voltage regulator that powers a noisy source, the interference will pop up again at the input of the voltage regulator. If we decouple the interference at the input of the voltage regulator, part of this interference pops back up at the voltage regulator output.
So the theoretically correct decoupling may not work that well in practice. Sure we measure lower ripple voltage on the active divider pin with a scope, but where did the ripple go? It simply moved to a different location where it could possibly do even more harm.
There are DIY members reporting improved sound quality when leaving out these decoupling caps altogether.
So what's the worse thing that could happen when we leave out all DEM caps? We end up with some radio frequency ripple on the output signal. This radio frequency can be dissipated by means of suitable lossy ferrite beads and we end up with the averaged current again.
If we increase the DEM clock frequency by using say 100pF (roughly 1 MHz DEM clock) timing cap, we could increase the ripple frequency so it could be dissipated more easily when using ferrite beads.
The ferrite beads could be placed between TDA1541A output and a passive I/V resistor or between the TDA1541A output and active I/V converter circuit.
Good morning,
John, I read the following on your webpage:
"Music
Music stored on NAS
Music is delivered to RPI3 from minimServer + minimStreamer (UPnP/DNLA)
Music is transcoded to wav to offload RPI3
Separate setting required in minimStreamer to support transcoding
Tested with Synology DS216e NAS"
Do you mean DS216se with its single core processor has enough power for transcoding audio data?
Ernst
John, I read the following on your webpage:
"Music
Music stored on NAS
Music is delivered to RPI3 from minimServer + minimStreamer (UPnP/DNLA)
Music is transcoded to wav to offload RPI3
Separate setting required in minimStreamer to support transcoding
Tested with Synology DS216e NAS"
Do you mean DS216se with its single core processor has enough power for transcoding audio data?
Ernst
Hi, I asked because I couldn't find any spec about the DS216e, only DS216se, which is the lowest end of the 216 versions. Are they camparable or is DS216e just a typo?
Best
Ernst
John, bedankt!
Ok, good to know!
Since I am not yet a happy owner of your well regarded Mosaic DACs I sometimes do play higher sampled material.
Do you accidently know, if the 216se is able to transcode upto 384 khz as well?
Best regards
Ernst
Because the sonons it produces are fatter
BTW, do you guys know this new method of CD production??
CDJapan : All About Ultimate High Quality CD
My first items are still traveling but online samples seem promising.
M.
Please check the whole collection. There are dozens of true gems, including many reisues of Furtwängler's most wanted...
Probably, but who cares, if it sounds more engaging than "no distortion at all" comming from non-Mosaic state of the art DACs...
I'm not John, but I am using a DS112J (Synology) and transcoding works without any issues too.
I've gone trough several UPnp control points (Android) and I've found some are much slower than others (Kazoo is slow and I dislike its functionality/UI). With the better ones, the DS112J does a good job as a server when playing music from the NAS (have not tried anything else). I am using Hi-Fi Cast at this point.
Last edited: