Madness of dropped samples and bits?
Here's a little contribution to the discussion from a recording engineer...
I've joined this forum because it seems to allow people to listen while always looking for the purely technical truth and explanation. It's a great combination and the only one that works if you ask me.
For this to work it's OK to have different opinions, but we should respect both the people who know the theory well and those with good ears. IWe need both. n the end, audio equipment should be well designed for listening to great music, right?
Now, to the point. I recently discovered something that really puzzled me. Working with a master for a CD-release I took the time to compare different types of dither and included a non-dithered, truncated version (20 to 16 bits in this case). To my big surprise, and without any doubt, the truncated was the winner, so I ended up dithering the fades only. This is against all I've learned about dithering... An explanation would be highly appreciated...
Without a proper listening test I would never rule out the idea of an occasional dropped sample compared to a jittery clock extraction. At the same time, I would still keep looking for a solution that includes all samples and sounds great.
Sorry if this is slightly OT and old news...
Martin
Here's a little contribution to the discussion from a recording engineer...
I've joined this forum because it seems to allow people to listen while always looking for the purely technical truth and explanation. It's a great combination and the only one that works if you ask me.
For this to work it's OK to have different opinions, but we should respect both the people who know the theory well and those with good ears. IWe need both. n the end, audio equipment should be well designed for listening to great music, right?
Now, to the point. I recently discovered something that really puzzled me. Working with a master for a CD-release I took the time to compare different types of dither and included a non-dithered, truncated version (20 to 16 bits in this case). To my big surprise, and without any doubt, the truncated was the winner, so I ended up dithering the fades only. This is against all I've learned about dithering... An explanation would be highly appreciated...
Without a proper listening test I would never rule out the idea of an occasional dropped sample compared to a jittery clock extraction. At the same time, I would still keep looking for a solution that includes all samples and sounds great.
Sorry if this is slightly OT and old news...
Martin
Why the endless speculation? For any pair of clocks it’s very easy to calculate the expected number of duplicated or dropped samples. I posted a spreadsheet to do just that some time ago. And it’s easy enough to modify a wave file to create as many duplicate or missing samples as you like and determine your own hearing threshold. I predict that anyone who likes the TDA154x will tolerate a very large number of missing or duplicate samples.
Re: Re: 8xNOSDAC
Yes but even the nutty-as-a-fruitcake hardcore Japanese DIYers understand the need to create the necessary conditions for linear interpolation to work well. All the designs I've seen in MJ have a 4x or 8x oversampling digital filter before the linear interpolation.
chengtaw said:
Yes but even the nutty-as-a-fruitcake hardcore Japanese DIYers understand the need to create the necessary conditions for linear interpolation to work well. All the designs I've seen in MJ have a 4x or 8x oversampling digital filter before the linear interpolation.
Slightly off topic experiment
Hi,
I have read this thread with great interest. I don't know if I prefer the sound of missing samples over jitter - to make that claim I have too little practical experience. I have however implemented a dac with cs8416->pcm1798->D1 I/V, and listened to it with some mixed feelings. It is definitely better than cheap CD-players, but it doesn't knock my socks off.
I connected the dac to my HTPC (computer close to the tv
) and got much better sound than from my onboard VIA AC97 soundcard. Then I suddenly got the idea that I should test the sampling rate capabilities of my dac and forced the pc to upsample a 44.1k/128bit radio stream, and I was quite surprised with the sound I got. 
88.2k sounded horrible - digital artifacts or distortion, but 96k sounded much nicer (warmer) than standard 44.1k. 192k sounded nice too. I ran a blind test with my girlfriend and she picked 96k as better than 44.1k without a doubt. So the question is what am I hearing:
Smoothing of a crippled (compressed) signal from a net radio stream?
Less jitter distortions at high clock rates?
Better performance of cs8416 or pcm1798 at dvd sampling rates?
Some mojo with the spdifsignal - reflections and impedance matching should be worse with higher clock rates.
Something else?
Normally, higher clock speeds make things more complicated, so I'm a bit surprised with what I'm hearing. I am planning to have another go at designing the digital stuff, maybe with another dac or receiver, but I'm beginning to think that the clock signals cannot be ignored - I thought it would be enough just to make short traces between the chips, but if the cs8416 chip is playing it own song, I guess I'll have to think of some sort of reclocking preferably adaptive.
BTW: The reason 88.2 sounded bad is probably because I was using a fast filter for upsampling - I simply ran mplayer with -srate 88200 as option.
Hi,
I have read this thread with great interest. I don't know if I prefer the sound of missing samples over jitter - to make that claim I have too little practical experience. I have however implemented a dac with cs8416->pcm1798->D1 I/V, and listened to it with some mixed feelings. It is definitely better than cheap CD-players, but it doesn't knock my socks off.
I connected the dac to my HTPC (computer close to the tv
) and got much better sound than from my onboard VIA AC97 soundcard. Then I suddenly got the idea that I should test the sampling rate capabilities of my dac and forced the pc to upsample a 44.1k/128bit radio stream, and I was quite surprised with the sound I got. 88.2k sounded horrible - digital artifacts or distortion, but 96k sounded much nicer (warmer) than standard 44.1k. 192k sounded nice too. I ran a blind test with my girlfriend and she picked 96k as better than 44.1k without a doubt. So the question is what am I hearing:
Smoothing of a crippled (compressed) signal from a net radio stream?
Less jitter distortions at high clock rates?
Better performance of cs8416 or pcm1798 at dvd sampling rates?
Some mojo with the spdifsignal - reflections and impedance matching should be worse with higher clock rates.
Something else?
Normally, higher clock speeds make things more complicated, so I'm a bit surprised with what I'm hearing. I am planning to have another go at designing the digital stuff, maybe with another dac or receiver, but I'm beginning to think that the clock signals cannot be ignored - I thought it would be enough just to make short traces between the chips, but if the cs8416 chip is playing it own song, I guess I'll have to think of some sort of reclocking preferably adaptive.
BTW: The reason 88.2 sounded bad is probably because I was using a fast filter for upsampling - I simply ran mplayer with -srate 88200 as option.
Ulas said:
The above paragraph pretty much describes the schematic. It had no part numbers and few, if any, logic gates, just block diagrams with the target frequencies. Come to think of it, it had no parts. It was more a synopsis than a finished script. Think of it as a seed which the more enterprising could grow into something of their own.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.