MDAC First Listen (part 00110010)
- Thread starter 44000khz
- Start date
- Status
All the best wishes for the trip John. How is the new clock coming up?So HK/China trip starts Oct 9th... lets see how Detox goes
DANOFDANGER
pfm Member
Hey John, i wonder if you can help me out with a troubling matter of mine. Dealing with the Vega i have noticed a upsetting issue. In my previous statement about the Vega, i detailed about how it had troubles with lower octave instruments compared to the Mdac. I started to investigate this and began experimenting on the problem. I started with the filters on the Vega and found that mode 4 was the cause (the filter i had chosen early on). So i investigated a little more and starting comparing filters with the Mdac. I soon discovered that minimum phase of the Mdac exhibited the exact same issues.
The problem i am experiencing i describe as best i can as a imbalance of resolution over a frequency response. What i mean is that lower frequency sounds have less detail compared to higher frequency. Its as if the resolution of the output changes depending on what the frequency is.
I found this issue was only with the Mdacs minimum phase and Vegas Mode 4, which i soon found out is a minimum phase as well.
All other filters did not have this problem.
So i spent the day researching about filters for d to a converters and come across something interesting. A article by Resonessnce Labs talking about filters and their impact on audio conversion today. Upon reading i found out about certain shortfalls of filters as they try to rectify the phenomenons created due to tight constraints of the filters. What was most interesting was with non linear phase filters and how they have problems with dispersion.
With non linear phase filters they have trouble with keeping group delays linear over different frequency's. This creates the problem described as dispersion. The article went on to describe in detail about what this might sound like, such as perceived loudness differences between instruments, unrealistic sounds, instrument notes distance differences. As sounds will arrive differently through the dispersive filter with different frequencies, this will be interpreted by the ear as a different distance to the instruments, and not as a distortion.
It made me wonder if this was in fact what i was hearing. Have you experienced anything similar ?
All i know now is that i can pick this up and find it more unsatisfying compared to linear filters. For sure it appears to be much nicer with higher frequency information but the drawbacks are too disturbing.
This leads me to some questions i have about filters and the Fdac that i hope you could answer.
As i understand it sigma deltas natural oversampling allows for a more simpler and cheaper reconstruction filter as the spectral images are shifted further from the passband. So i can only assume that the digital filters are either anti aliasing filters for different images introduced because of the nature of SD modulators. Or they are just simply HF filters because of the noise generated due to noise shaping and therefore must be filtered out. But why ?
Also why does digital filtering remove the benefit of oversampling. Of all the frequency plots i have seen, the filters appear to be constraint to the bandwidth of the sample rate used (for eg 44.1khz) like that of conventional converters without oversampling. Is this why your optimal transient filters remove all pre and post ringing ? because you allow a much nicer roll off compared to fast roll off filters.
Now regarding the Fdac, how are you intending to implement your filters ?. Are you still going to be focusing on your optimal transient as the golden standard or do you have something else in mind. It has made me quite concerned about the very limitations of PCM at 44.1khz. Is there any way we can create a filter that exhibits a frequency plot like that of a typical fast roll off, but also has good time domain response like your optimum transient. To me the closest we could possibly get would be your optimal transient at 88.2ksps with the roll off starting above 20khz. Or perhaps is it possible to combine the benefits of linear and non linear phase filters to gain a best of both worlds type filter, this i would be most interested to see.
I think you have ruined me for life with your optimal transient filters however..
This has made me think that this is perhaps why DSD is better, because of the absence of ringing during recording and playback.
Damn you SONY.
I understand your busy, i always try to research as much as i can before i have to come to you. But when things as troubling as this arrive, getting your insight is very very helpful.
Thanks.
The problem i am experiencing i describe as best i can as a imbalance of resolution over a frequency response. What i mean is that lower frequency sounds have less detail compared to higher frequency. Its as if the resolution of the output changes depending on what the frequency is.
I found this issue was only with the Mdacs minimum phase and Vegas Mode 4, which i soon found out is a minimum phase as well.
All other filters did not have this problem.
So i spent the day researching about filters for d to a converters and come across something interesting. A article by Resonessnce Labs talking about filters and their impact on audio conversion today. Upon reading i found out about certain shortfalls of filters as they try to rectify the phenomenons created due to tight constraints of the filters. What was most interesting was with non linear phase filters and how they have problems with dispersion.
With non linear phase filters they have trouble with keeping group delays linear over different frequency's. This creates the problem described as dispersion. The article went on to describe in detail about what this might sound like, such as perceived loudness differences between instruments, unrealistic sounds, instrument notes distance differences. As sounds will arrive differently through the dispersive filter with different frequencies, this will be interpreted by the ear as a different distance to the instruments, and not as a distortion.
It made me wonder if this was in fact what i was hearing. Have you experienced anything similar ?
All i know now is that i can pick this up and find it more unsatisfying compared to linear filters. For sure it appears to be much nicer with higher frequency information but the drawbacks are too disturbing.
This leads me to some questions i have about filters and the Fdac that i hope you could answer.
As i understand it sigma deltas natural oversampling allows for a more simpler and cheaper reconstruction filter as the spectral images are shifted further from the passband. So i can only assume that the digital filters are either anti aliasing filters for different images introduced because of the nature of SD modulators. Or they are just simply HF filters because of the noise generated due to noise shaping and therefore must be filtered out. But why ?
Also why does digital filtering remove the benefit of oversampling. Of all the frequency plots i have seen, the filters appear to be constraint to the bandwidth of the sample rate used (for eg 44.1khz) like that of conventional converters without oversampling. Is this why your optimal transient filters remove all pre and post ringing ? because you allow a much nicer roll off compared to fast roll off filters.
Now regarding the Fdac, how are you intending to implement your filters ?. Are you still going to be focusing on your optimal transient as the golden standard or do you have something else in mind. It has made me quite concerned about the very limitations of PCM at 44.1khz. Is there any way we can create a filter that exhibits a frequency plot like that of a typical fast roll off, but also has good time domain response like your optimum transient. To me the closest we could possibly get would be your optimal transient at 88.2ksps with the roll off starting above 20khz. Or perhaps is it possible to combine the benefits of linear and non linear phase filters to gain a best of both worlds type filter, this i would be most interested to see.
I think you have ruined me for life with your optimal transient filters however..
This has made me think that this is perhaps why DSD is better, because of the absence of ringing during recording and playback.
Damn you SONY.
I understand your busy, i always try to research as much as i can before i have to come to you. But when things as troubling as this arrive, getting your insight is very very helpful.
Thanks.
adamdea
You are not a sound quality evaluation device
I'm not John obviously, and not an expert either, but I may be able to help with one bit: the digital antialias (A/D) and anti imaging (D/A) filters appear after delta sigma modulation (A/D) and before the delta sigma modulation (D/A) . This is the basis of the Sony diagram justifying DSD as a playback mechanism- it misses out both of those filters.
Also note that oversampling and "digital filter" are pretty much interchangeable concepts since (as follows from the sampling theorem) the only way to have digital data free of the spectral images implicit in the samples is to convert to a higher sample rate: there is no way of expressing 16/44 data without the spectral images in 16/44; the filter must output a higher sample rate data.
So the input of the delta sigma modulator after the AI filter will (I think) be more than 16/44.
If you omit an AA filter when replaying 16/44 the delta sigma modulator will faithfully reproduce the spectral images implicit in the samples. [as I understand from what John has said the OT filters upsample by repeating the data values ie effectively reproducing the effect of a zero order hold dac; this is rather like not having a digital filter at all -but there seems to be a bit more to it than that as DD and XD are different]
Also you might want to consdier the MQA documentation for the B spline kernel and even linear interpolation (first order hold) for an example of a filter which has some HF roll off and good time domain behaviour (if you believe in that stuff- but note that the concept is not used with any fixed meaning : Robert Watts uses it to describe the behaviour of a filter which rings til kingdom comes).
The MQA blurb is specifically addressed at achieving a novel balance between frequency and time domain response. Definitely worth a read (with the caution: "this may contain marketing material").
DANOFDANGER
pfm Member
Any help is appreciated. Yes this is what i thought, because of the abscence of these filters the pre and post ringing would be non existent. As the filters themselves are the cause of said effects.
I am not sure what you mean by oversampling and digital filters used interchangably, although oversampling can result in filter characteristics i would not myself use it when talking about filters or vice versa. But yes the spectral images are always there and even when filters are applied during the A to D process, the images cannot be removed as its not something embedded in the sampling process but something in which is always there due to the nature described by the sampling theorem.
But oversampling does not remove spectral images, it only shifts them further away from the passband so that it cannot cause aliasing or filters can simply wipe it away from care. This was one big aspect of oversampling and conventional converters as it made possible for analog filters to be made with less complexity and higher linearity. But the question differs. Because S D modulators oversample to improve SNR and linearity, they must naturally push spectral images far outside the band of concern. So at the very least, only a simple analog reconstruction filter is used. But the digital filters placed must be for another reason. Of one i can only think of is modulation noise from the converter just outside the passband that must be delt with. For what reason i dont know other than it somehow affecting the passband.
If i somehow missed something please excuse my ignorance.
Im not sure if i am lost but the AA and AI is always after the multibit modulator in both A to D and D to A conversion...
I think you would have ment oversampling here as upsampling is to do with sampling output rate from a source. But as far as i was aware, optimal transient was of a non oversampling type. DD and XD are aparently the same filter driven by different mathematical calculations to arrive at the same value.
If you could supply a link il be sure to have a look. It all depends if i can interperate it however as my understanding of digital filters falls off sharply with involved math theory.
Have you heard much about apodizing filters ? As i have read they claim to provide linear phase filters with near perfect pre ringing abscence with added post ringing. Its not something i see a lot of.
JohnW
pfm member
Yes, Correct.
Again correct, the OT "Filters" are simply sample and hold... and as such introduces no Digital domain* pre or post ringing on the playback side. * the analogue domain has a small amount of overshoot / undershoot.
https://dl.dropboxusercontent.com/u/86116171/MDAC Optimal Transient.jpg
There is no clear differentiation between minimum phase and apodizing filters - apodizing filters are more marketing speak and just a variation on minimum phase. (as far as I understand the apodizing filter achieves its "full" digital attenuation at FS/2 (as Shannon theory demands) rather then just 6dB of typically poorly designed digital filters, but this is not a restriction of "minimum phase"....
adamdea
You are not a sound quality evaluation device
I just meant that a digital filter needs to output a higher sample rate in order to filter. If you understand the relationship between sample rate and spectral images (this is stated clearly in Shannon's proof of the sampling theorem) then this is clear.
An oversampling filter does remove spectral images only not all of them. The number of them removed is determined by the output sample rate, for reasons which are obvious once one get the connection between sample rate, nyquist and spectral images
no. Oversampling is necessary for a digital AA/AI filter but not sufficient. [incidentally I think the OT filters prove this since they oversample without filtering]The SD modulators in a DAC will faithfully reproduce what they are given. I think you might be mixing up A/D and d/a. SD dac chips have the advantage of improving linearity and resolution. I don't think it's to do with filtering. in an A/D process they do of course assist with filtering
Don't think so: it's before the SD modulator in the D/A process. Have a look at the famous Sony why DSD is great graphic if you don't believe me.
[edit I'm talking about digital filters- there will be an analog AI/delta sigma noise filter at the very end]
There are 3 of them. Two of them are basically the same, the other isn't because it doesn't produce weird distortions in the audible band (at least that's what BE718 measured)
Sorry I do not have them to hand. There have been loads of threads on this forum with the relevant links.Really you can't miss it. EDIT Actually try this http://www.aes.org/tmpFiles/elib/20160913/17501.pdf
Apodising seems to be used variously to refer to two different concepts which accordingly get mixed up. One is a filter which cuts in under the lower frequency limit of the preceding filters and therefore removes the ringing embedded by them. The other is a minimum phase filter characteristic of a filter which only post-rings. If you put them together you have a filter which removes the ringing embedded in the A/D process and replaces it with post ringing.
adamdea
You are not a sound quality evaluation device
If I remember our discussion about this ages ago, I think you said that OT does upsample 16/44 it just doesn't filter. I understood this to mean that the sample values are repeated so as to produce sample and hold in the digital domain (appreciate it might be a bit more complicated). Did I get that wrong?
JohnW
pfm member
Any news on the prototype?
Michael,
This weekend we worked on the OT filters for the DAC board, the NEW ESS Dac's have a different filter structure to the older ESS9018 (we have a temporary solution working but they overload at 0dB and 20KHz, but OK for listening tests) - OK until we can get more info from ESS.
Tomorrow, Wednesday we get a new power-amp as we killed one of the modified 8200M's driving the CLS's
so we will have our "first" listening tests of the "prototype" DAC PCB tomorrow.Hi John
I bought a Schiit Modi Multibit. Music is a bit less "scratchy" and harsh now compared with the MDAC but it hasn't totally solved the problem. Therefore, I think the sonic problems I have been experiencing are a combination of the MDAC and my Focal CMS 50 speakers (they have a metal tweeter). I think it is known as bad synergy.
There is no big hurry to get my MDAC repaired, now that I have the Schiit DAC to keep me going. I know you are very busy. But it would be nice to have it back by Christmas, if possible
Best wishes
mrflibble
I bought a Schiit Modi Multibit. Music is a bit less "scratchy" and harsh now compared with the MDAC but it hasn't totally solved the problem. Therefore, I think the sonic problems I have been experiencing are a combination of the MDAC and my Focal CMS 50 speakers (they have a metal tweeter). I think it is known as bad synergy.
There is no big hurry to get my MDAC repaired, now that I have the Schiit DAC to keep me going. I know you are very busy. But it would be nice to have it back by Christmas, if possible
Best wishes
mrflibble
adamdea
You are not a sound quality evaluation device
Do you think that (later on) you might have a go at an MQA type filter. Obviously the precise details are a closely guarded secret etc but the general gist seems to be explained in the documentation.Michael,
This weekend we worked on the OT filters for the DAC board, the NEW ESS Dcc's have a different filter structure to the older ESS9018 (we have a temporary solution working but they overload at 0dB and 20KHz, but OK for listening tests) - OK until we can get more info from ESS.
Tomorrow, Wednesday we get a new power-amp as we killed one of the modified 8200M's driving the CLS's
mrflibble
By then we should have the MDAC2 PCB's
@John,
Sorry to persist on this.
I have no recollection of us all agreeing on a change in the project schedule i.e the FDAC having first priority.
I think it's a bit unfair vs us supporters if you, as it seems, have changed the schedule with MDAC2 (awakenig from the dead) taking priority over the FDAC.
There's always consequences of decisions being made. One consequence for me of the joint decision we all made to abandon the MDAC2 in favor of the FDAC was that I sold my MDAC ...
I do have a habit of trusting decisions made./Lars
PS: I'd appreciate if only John responds to this ...
DS
JohnW
pfm member
Lars,
I'm working on the FDAC - but I'm very keen to get a design out this side of Christmas.
With the MDAC2 now "basically" developed Jarek can combine the Superior PCB with our XMOS development board and - due to the advance stage of development this can be completed within a couple of weeks and as the PCB will be made in Europe allowing MDAC2 shipped well in time for Christmas.
I hope with the Detox and MDAC2 completed before the end of the year this will allow time for the FDAC to be completed in due coarse.
Over the summer period I have made connections with production here in Czech Rep. and this makes the MDAC2 even more practical.
There was no direct intention to have the MDAC2 ready ahead of the FDAC, its just happen as a natural "development" of the FDAC...
I see many people being VERY happy with the MDAC2 - certainly with the superior connected to the MDAC it sounds very very good and we will listen to our XMOS solution with Fusions superior board tomorrow - but even with the XMOS's simple analogue stage it sounds surprisingly good
I feel its VERY important to get a couple of designs completed and shipped THIS year... the MDAC2 is a simpler task then the FDAC.
I hope that we can have MDAC2 into prototype production while I'm in China next month so I can confirm upon my return then we are 1 month or so from mass production.
I still need to solve the external PSU for the MDAC2 - I might find a solution during the China / HK trip.
I'm working on the FDAC - but I'm very keen to get a design out this side of Christmas.
With the MDAC2 now "basically" developed Jarek can combine the Superior PCB with our XMOS development board and - due to the advance stage of development this can be completed within a couple of weeks and as the PCB will be made in Europe allowing MDAC2 shipped well in time for Christmas.
I hope with the Detox and MDAC2 completed before the end of the year this will allow time for the FDAC to be completed in due coarse.
Over the summer period I have made connections with production here in Czech Rep. and this makes the MDAC2 even more practical.
There was no direct intention to have the MDAC2 ready ahead of the FDAC, its just happen as a natural "development" of the FDAC...
I see many people being VERY happy with the MDAC2 - certainly with the superior connected to the MDAC it sounds very very good and we will listen to our XMOS solution with Fusions superior board tomorrow - but even with the XMOS's simple analogue stage it sounds surprisingly good
I feel its VERY important to get a couple of designs completed and shipped THIS year... the MDAC2 is a simpler task then the FDAC.
I hope that we can have MDAC2 into prototype production while I'm in China next month so I can confirm upon my return then we are 1 month or so from mass production.
I still need to solve the external PSU for the MDAC2 - I might find a solution during the China / HK trip.
JohnW
pfm member
Reading between the lines the MQA filter is a combination of the optimal spectrum (full attenuation between 20KHz and FS/2) and minimum phase - so its nothing special.
russel
./_dazed_and_confused
Hi John
I bought a Schiit Modi Multibit. Music is a bit less "scratchy" and harsh now compared with the MDAC but it hasn't totally solved the problem. Therefore, I think the sonic problems I have been experiencing are a combination of the MDAC and my Focal CMS 50 speakers (they have a metal tweeter). I think it is known as bad synergy.
There is no big hurry to get my MDAC repaired, now that I have the Schiit DAC to keep me going. I know you are very busy. But it would be nice to have it back by Christmas, if possible
Best wishes
mrflibble
If you have a metal tweeter then it may accentuate switching distortion, I use a current dumping design which has a much sweeter sound, class A like. It's based on the Quad 405 except it has modern complementary output transistors and modern low distortion op-amp. If you want the design files, you are welcome, as is anyone else, it will most likely tide people over until John produces his own error correcting design.
adamdea
You are not a sound quality evaluation device
I quite agree that it's a good idea to get something finished. But this is the first time I can remember it has been suggested that the F dac would not be complete this year. Next year sounds a long long way away.Lars,
I'm working on the FDAC - but I'm very keen to get a design out this side of Christmas.
With the MDAC2 now "basically" developed Jarek can combine the Superior PCB with our XMOS development board and - due to the advance stage of development this can be completed within a couple of weeks and as the PCB will be made in Europe allowing MDAC2 shipped well in time for Christmas.
I hope with the Detox and MDAC2 completed before the end of the year this will allow time for the FDAC to be completed in due coarse.
Over the summer period I have made connections with production here in Czech Rep. and this makes the MDAC2 even more practical.
There was no direct intention to have the MDAC2 ready ahead of the FDAC, its just happen as a natural "development" of the FDAC...
I see many people being VERY happy with the MDAC2 - certainly with the superior connected to the MDAC it sounds very very good and we will listen to our XMOS solution with Fusions superior board tomorrow - but even with the XMOS's simple analogue stage it sounds surprisingly good
I feel its VERY important to get a couple of designs completed and shipped THIS year... the MDAC2 is a simpler task then the FDAC.
I hope that we can have MDAC2 into prototype production while I'm in China next month so I can confirm upon my return then we are 1 month or so from mass production.
I still need to solve the external PSU for the MDAC2 - I might find a solution during the China / HK trip.
I think that it would be sensible to create a proper real achievable strict road map for the Fdac and to allow people the option to transfer the payments made for the FDac into mdac2s. Maybe the mdac 2 just is the mdac replacement we've always wanted. I was excited at the idea of the complete ADC\dac phono streamer but maybe I'll have to settle for less.
DANOFDANGER
pfm Member
Yea i get you, oversampling is needed in order to obtain the possibility of digital filtering.
I am sure you mean the OT oversample without "digital filters" they still indeed filter though, from the frequency plots of the filters they show all low pass filtering inside the passband. Utilizing the passband instead of leaving it mostly intact like most other filters. Its essentially sacrificing passband linearity for time domain linearity. I am not even sure it does not use digital filters, i think it still does.
When oversampling came along in conventional r2r dacs, it pushed sampling images further away from the passband. Allowing a more simpler anti imaging analog filter to be user. When sigma delta arrived, it just did this exessively more. When they changed the LPF to digital this allowed the analog filter to act only as a reconstruction filter, allowing it to be a simple rc circuit rather than a multi order filter for the 20-24khz band.
I think those dsd images have the filter before the SD modulator to show how it is bypassed, if they had the filter after they could not demonstrate their point as the line would show it bypassing the SD modulator as well.Don't think so: it's before the SD modulator in the D/A process. Have a look at the famous Sony why DSD is great graphic if you don't believe me.
[edit I'm talking about digital filters- there will be an analog AI/delta sigma noise filter at the very end]
Flick through this white paper for me. I have seen a few others with them all showing the digital filter after the modulator.
https://www.google.co.nz/url?sa=t&source=web&rct=j&url=http://www.analog.com/media/en/technical-documentation/application-notes/292524291525717245054923680458171AN283.pdf&ved=0ahUKEwjjm-TDj43PAhUEHpQKHdRiBWEQFgggMAI&usg=AFQjCNFum4gPsrSfpi3gNtaOsMhe_iswBg&sig2=_Nta0VYobAL1ya-oQn1Ozw
*EDIT*. After thoroughly looking at multiple papers i have concluded that the AA\AI filter is before ADC and "AFTER" DAC. The anti aliasing and anti imaging filters are handled by the analog LPF stages not the digital LPF. As far as i can tell, oversampling does not remove the spectral images of the source, but merely moves them further from the passband. I can understand that digital filters may remove unwanted sampling images, but they dont appear inside its filter bandwidth ?. Also regarding the digital low pass filter, i assume the filter is done in the oversampler/interploation modulator. Please confirm.
John has said they are all the same but with mathematical differences. Also from memory john could measure no differences in time or frequency domain between the three filters, i could be wrong though.
Thanks for that link adam, it looks very very detailed. I will do some reading and get back to you on all this. Thanks again.
- Status