'Chibi' phase inverting mod
Hi, Guido and I have discussed this issue before, two years ago, here:
http://db.audioasylum.com/cgi/m.mpl?forum=tweaks&n=60881&highlight=elso+guido&session=
Hi, Guido and I have discussed this issue before, two years ago, here:
http://db.audioasylum.com/cgi/m.mpl?forum=tweaks&n=60881&highlight=elso+guido&session=
Re: 'Chibi' phase inverting mod
And it wasn't resolved then either.
Elso Kwak said:Hi, Guido and I have discussed this issue before, two years ago, here:
http://db.audioasylum.com/cgi/m.mpl?forum=tweaks&n=60881&highlight=elso+guido&session=
And it wasn't resolved then either.
Re: Re: 'Chibi' phase inverting mod
The unequality between the two DACs for the positive and negative signal is much greater than the difference due the 1 least significant bit. You can check this out by submitting the two signals in phase to the balanced to single ended converter. You can play a little by making the IV resistor variable but this will also alter the frequency response.
I did not hear a significant difference between balanced or normal operation so I abandoned the idea.
The AD1865 was more suitable as you get the balanced signal from one chip. But I like the TDA1543 more.
What, It was resolved.rfbrw said:
The unequality between the two DACs for the positive and negative signal is much greater than the difference due the 1 least significant bit. You can check this out by submitting the two signals in phase to the balanced to single ended converter. You can play a little by making the IV resistor variable but this will also alter the frequency response.
I did not hear a significant difference between balanced or normal operation so I abandoned the idea.
The AD1865 was more suitable as you get the balanced signal from one chip. But I like the TDA1543 more.
Re: Re: Re: 'Chibi' phase inverting mod
All you seem to be doing is testing the output stage. The nature of the output stage does not determine the matching between the dacs. As for the rest opinions vary. I can't stand the TDA1543.
Elso Kwak said:
What, It was resolved.
The unequality between the two DACs for the positive and negative signal is much greater than the difference due the 1 least significant bit. You can check this out by submitting the two signals in phase to the balanced to single ended converter. You can play a little by making the IV resistor variable but this will also alter the frequency response.
I did not hear a significant difference between balanced or normal operation so I abandoned the idea.
The AD1865 was more suitable as you get the balanced signal from one chip. But I like the TDA1543 more.
All you seem to be doing is testing the output stage. The nature of the output stage does not determine the matching between the dacs. As for the rest opinions vary. I can't stand the TDA1543.
Re: Re: Re: 'Chibi' phase inverting mod
In those days, i did not see the light yet and was trying to not invert the LSB. The TDA1541 has -2mA output at digital zero.
So the stage after it has to cope with -2mA DC at rest. Full scale is from 0 to -4mA. So what happens as we invert data: the output will be -2mA+the nA current of the LSB (neg value).
Wont make one bit of a difference for the analogue stage.
Using a i/v resistor there will just be a bit more dc over it, referenced to yes the dacs agnd. So what? Any non DC signal coming out will be fine.
One thing which could count: if the dac is designed to have the least ammount of thd or non-linarity or whatever at the changes from 0 to -1 or +1, it might have a bit more at the change from -1
to 0 or -2.
For me, i'm using two i/v resistors to gnd and the a transformer over them, so from dac output to dac output. This cancels out the dc over the resistors coming from -2mA*R apart from the lsb difference. Not afraid of that being over the transformers input.
And the dc is not amplified by 15 by the transformer (as the signals are)
Sorry i don't have a nice dual scope to prove my point, i'll try to use the soundcard for this and will post! If i'm wrong, i'll change my dac design
Elso Kwak said:
What, It was resolved.
The unequality between the two DACs for the positive and negative signal is much greater than the difference due the 1 least significant bit. You can check this out by submitting the two signals in phase to the balanced to single ended converter. You can play a little by making the IV resistor variable but this will also alter the frequency response.
I did not hear a significant difference between balanced or normal operation so I abandoned the idea.
The AD1865 was more suitable as you get the balanced signal from one chip. But I like the TDA1543 more.
In those days, i did not see the light yet
and was trying to not invert the LSB. The TDA1541 has -2mA output at digital zero.So the stage after it has to cope with -2mA DC at rest. Full scale is from 0 to -4mA. So what happens as we invert data: the output will be -2mA+the nA current of the LSB (neg value).
Wont make one bit of a difference for the analogue stage.
Using a i/v resistor there will just be a bit more dc over it, referenced to yes the dacs agnd. So what? Any non DC signal coming out will be fine.
One thing which could count: if the dac is designed to have the least ammount of thd or non-linarity or whatever at the changes from 0 to -1 or +1, it might have a bit more at the change from -1
to 0 or -2.
For me, i'm using two i/v resistors to gnd and the a transformer over them, so from dac output to dac output. This cancels out the dc over the resistors coming from -2mA*R apart from the lsb difference. Not afraid of that being over the transformers input.
And the dc is not amplified by 15 by the transformer (as the signals are)
Sorry i don't have a nice dual scope to prove my point, i'll try to use the soundcard for this and will post! If i'm wrong, i'll change my dac design
It si very simple. As far as the dac is concerned nothing happens when you invert the data. 0000 0000 0000 1111 or any other value will continue to give out the same voltage as it always did irrespective of what happened to it prior to arriving at the dac. The idea that you can move bipolar zero is complete nonsense.
rfbrw said:
Okay - last try ...
1. Why not ? I do that (not for audio-purposes)
2. You know that a DAC has most distortions around bipolar zero, because all bits switching ? (Except 1Bit-Noiseshaper) That's why BurrBrown move this point to the end of the range (less distortions) and spend a second DAC. One for positive (0-data at bipolar zero) and one for negative (1-data at bipolar zero) half-wave. See Datasheet for the PCM63.
http://focus.ti.com/lit/ds/symlink/pcm63.pdf
3. I think you shouldn't care about DC in Audio - DC is the easiest think to remove !
4. You calculate DC in dB
If -1 isn't zero - how much dB is the step from 0 to 1 ? (on the positive DAC this is 1 to 2) ;-)
5. Independent of that: In case of a 20-Bit DAC changing 1LSB equals -120dB ! (all amps in the signal-path have more offset)
Regards
Jobstens
jobstens said:
Okay - last try ...
1. Why not ? I do that (not for audio-purposes)
2. You know that a DAC has most distortions around bipolar zero, because all bits switching ? (Except 1Bit-Noiseshaper) That's why BurrBrown move this point to the end of the range (less distortions) and spend a second DAC. One for positive (0-data at bipolar zero) and one for negative (1-data at bipolar zero) half-wave. See Datasheet for the PCM63.
http://focus.ti.com/lit/ds/symlink/pcm63.pdf
3. I think you shouldn't care about DC in Audio - DC is the easiest think to remove !
4. You calculate DC in dB
If -1 isn't zero - how much dB is the step from 0 to 1 ? (on the positive DAC this is 1 to 2) ;-)
5. Independent of that: In case of a 20-Bit DAC changing 1LSB equals -120dB ! (all amps in the signal-path have more offset)
Regards
Jobstens
Table2 on page 7 tells you all you need to know. There is a chart and that chart lays down how much current will be dumped into a virtual earth for each value between plus and minus fullscale. If you can't understand such simplicity, then there is nothing else to be said.
Sure - that's the result. But you say that bipolar zero is fixed in a DAC. In this DAC are two DACs. One is 1111111111111111111 and the other is 0000000000000000000 at bipolar zero. That's the point.
You can take two DACs, connect both current-outputs and design a logic that feed them like in PCM63. Now you have moved your zero-point for each DAC.
In our phase-inverter-circuit you also could simply invert the signal (with a single inverter) and, if you need, add the 1 LSB with a simple resistor. You now have moved the zero-point by 1 LSB.
Maybe you should build a converter (for playing purposes) with an adder (e.g. 4008 or 7483 - you need 4 for 16 Bit) in front of the DAC with DIP-switches on the second input. But you will have a Problem at overflow ... (you could also burn calculated values in an eprom)
I design Audio-Circuits since 26 years - I know that it is like I tell you ...
(I don't expect that you believe me, but I'll not answer anymore - your learn-resistance is too much for me)
Regards
Jobstens
You can take two DACs, connect both current-outputs and design a logic that feed them like in PCM63. Now you have moved your zero-point for each DAC.
In our phase-inverter-circuit you also could simply invert the signal (with a single inverter) and, if you need, add the 1 LSB with a simple resistor. You now have moved the zero-point by 1 LSB.
Maybe you should build a converter (for playing purposes) with an adder (e.g. 4008 or 7483 - you need 4 for 16 Bit) in front of the DAC with DIP-switches on the second input. But you will have a Problem at overflow ... (you could also burn calculated values in an eprom)
I design Audio-Circuits since 26 years - I know that it is like I tell you ...
(I don't expect that you believe me, but I'll not answer anymore - your learn-resistance is too much for me)
Regards
Jobstens
CS4390 datasheet:
while Format 6 is 24-bit I2S. 20.18. or 16-bit I2S can be implemented in Format 6 if the data is followed by 4. 6. or 8 zeros respectively to simulate 24-bit input as shown in figure 7. A very small offset will result if the 20. 18. or 16-bit data is followed by
static non-zero data.
Keywords offset and static non-zero data
while Format 6 is 24-bit I2S. 20.18. or 16-bit I2S can be implemented in Format 6 if the data is followed by 4. 6. or 8 zeros respectively to simulate 24-bit input as shown in figure 7. A very small offset will result if the 20. 18. or 16-bit data is followed by
static non-zero data.
Keywords offset and static non-zero data
You have a bad memory, suggest you re-read this thread again!
Discussion was if inverting I2S and not adding the LSB can be seen as DC offset or that it is (noticeble) distortion.
Here we go again: inverting two's complement:
step 1: invert all bits, step 2: add one.
If you don't do the second step (just inverting the dataline) you feed the dac with values which are 1 (lsb) too low. My view: dont bother, it's very little dc offset.
Now the datasheet. It's a 24bit dac, mode 6 is 24bit i2s feed.
You can feed it in this mode with 16 bit data if you leave the last 8 bits filled with 00000000. If you have static non-zero data, instead of zero's for the last 8 bits, the datasheet says it is dc offset. One options for static non-zero data is 00000001. So all the samples have the lsb set. So you are feeding the dac with values which are 1 (lsb) to high. Crystals view: it is a little offset.
I and others tried to explain why nobody bothers with the lsb error when inverting i2s since it is little dc. I don't expect that anyone believes some diy guys, but how about crystal
Discussion was if inverting I2S and not adding the LSB can be seen as DC offset or that it is (noticeble) distortion.
Here we go again: inverting two's complement:
step 1: invert all bits, step 2: add one.
If you don't do the second step (just inverting the dataline) you feed the dac with values which are 1 (lsb) too low. My view: dont bother, it's very little dc offset.
Now the datasheet. It's a 24bit dac, mode 6 is 24bit i2s feed.
You can feed it in this mode with 16 bit data if you leave the last 8 bits filled with 00000000. If you have static non-zero data, instead of zero's for the last 8 bits, the datasheet says it is dc offset. One options for static non-zero data is 00000001. So all the samples have the lsb set. So you are feeding the dac with values which are 1 (lsb) to high. Crystals view: it is a little offset.
I and others tried to explain why nobody bothers with the lsb error when inverting i2s since it is little dc. I don't expect that anyone believes some diy guys, but how about crystal
guido said:
Not from me it wasn't. Mine was purely a technical point namely that if you do not add the 1 then two's comp inversion is incorrect and you have distortion, an error, an offset or whatever you choose to call it and it SHOULD NOT BE THERE. As far as I am concerned if it should not be there it is distortion. Quote who you like but when the signal is supposed to be 1111 0000 and it comes out as 1111 0001 then to me it is distorted. A error is an error wherever it occurs. Its impact and whether or not it matters is something I would leave to others.
rfbrw said:
So now it is suddenly a purely technical point. Yeah right:
"Considering the serial nature of modern dacs it isn't really surprising that most don't bother with 1 bit addition."
"However, I do think if one has gone to the bother of creating a balanced dac, e.g. as shown in the AD1852 datasheet, the 1 bit addition after inversion should be mandatory."
Call it distortion or whatever you like. In the real world it is DC and that is why nobody (including audiomanufacturers from philips to pass) bothers to fix the LSB. You wont here it and it is only a waste of effort.
That should be the message here to anyone going with a balanced diy design (allthough i can imagine that somebody would address the lsb, just for the fun designing/building it).
As for the commerical guys. They are not going to: $$$ wasted.
If you call it distortion the newbee's would go
by just the 'd' word.
IT IS NOT £$&%&^G DC
You can wail all you like. To be technically correct you MUST add the 1. And if you go to the extent of creating a balanced dac, then technical accuracy should be paramount because incorrect inversion impacts on CMRR in the form of an inbuilt error relative to the uninverted datastream. I fully understand why no one bothers with the 1 addition for simple phase inversion but that does not stop it being an error.
The message, if any, is that it is a judgement call. At 24bits it probably doesn't matter and at 16bits Jbokelmann thinks you can hear the difference and I'd play safe and add the one.
Simple inversion uses an inverter as in 74/4000 logic. Unless you have other uses for a DSP you do not use one just to add one when considerably cheaper programmable logic is available.
Oh and one more thing. IT IS NOT DC. The data being inverted is dynamically varying. It is audio data. The inverse of dynamically varying audio data is inverted dynamically varying audio data. Why you think if you invert data the LSB it somehow sticks at one when the uninverted LSB was can also be a one is simply a mystery to me. It is inconceivable that the error can be static if the data prior to inversion was not static.
If you really cannot grasp the concept that the LSB varies along with the entire word from sample to sample then there really isn't much more I can say.
guido said:
So now it is suddenly a purely technical point. Yeah right:
"Considering the serial nature of modern dacs it isn't really surprising that most don't bother with 1 bit addition."
"However, I do think if one has gone to the bother of creating a balanced dac, e.g. as shown in the AD1852 datasheet, the 1 bit addition after inversion should be mandatory."
Call it distortion or whatever you like. In the real world it is DC and that is why nobody (including audiomanufacturers from philips to pass) bothers to fix the LSB. You wont here it and it is only a waste of effort.
That should be the message here to anyone going with a balanced diy design (allthough i can imagine that somebody would address the lsb, just for the fun designing/building it).
As for the commerical guys. They are not going to: $$$ wasted.
If you call it distortion the newbee's would go
You can wail all you like. To be technically correct you MUST add the 1. And if you go to the extent of creating a balanced dac, then technical accuracy should be paramount because incorrect inversion impacts on CMRR in the form of an inbuilt error relative to the uninverted datastream. I fully understand why no one bothers with the 1 addition for simple phase inversion but that does not stop it being an error.
The message, if any, is that it is a judgement call. At 24bits it probably doesn't matter and at 16bits Jbokelmann thinks you can hear the difference and I'd play safe and add the one.
Simple inversion uses an inverter as in 74/4000 logic. Unless you have other uses for a DSP you do not use one just to add one when considerably cheaper programmable logic is available.
Oh and one more thing. IT IS NOT DC. The data being inverted is dynamically varying. It is audio data. The inverse of dynamically varying audio data is inverted dynamically varying audio data. Why you think if you invert data the LSB it somehow sticks at one when the uninverted LSB was can also be a one is simply a mystery to me. It is inconceivable that the error can be static if the data prior to inversion was not static.
If you really cannot grasp the concept that the LSB varies along with the entire word from sample to sample then there really isn't much more I can say.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.