It's at the output of the DAC. If the signal is clipped before the SRC then it will be clipped after the SRC because the S/PDIF cannot handle anything above FS. I noticed that if I force the output of the sound-blaster to re-sample at 96kHz then neither signal is clipped.
cheers
The signal is not actually clipped. No values abow 1 or -1!
A perfect dac should output a pure sine at 1.08 volts if a sin at 0dBFS outputs 1 volt.
If you do a FFT on the signal in audacity it is a pure sine.
The magic is in the phase shift so the sine is never sampled at the max values of the sine. But it should of course be output as a sine curve, ref Nyquist.
Such a sine is difficult to sample in a ADC, but as the tc electonics paper shows, it is a common signal after modern mixing and mastering sound prosessing in the digital domain.
A perfect dac should output a pure sine at 1.08 volts if a sin at 0dBFS outputs 1 volt.
If you do a FFT on the signal in audacity it is a pure sine.
The magic is in the phase shift so the sine is never sampled at the max values of the sine. But it should of course be output as a sine curve, ref Nyquist.
Such a sine is difficult to sample in a ADC, but as the tc electonics paper shows, it is a common signal after modern mixing and mastering sound prosessing in the digital domain.
Last edited:
Well, you don't need to worry about the 96kHz sample rate on the MiniDSP... because it's actually 48kHz.
There's always the 4x10 HD though, which is 96 (and £500)All the more reason why getting something like this would be wonderful. It could be more than the Hypex DSP promised (and failed) to be...
So as I anciently await the completion of this project to implement it in my system as an active 4 way crossover, or at least to start.
So at the risk of hi-jacking this thread, I propose a question. I am trying to understand why one would have the need for a word length of 192khz or better. If you can achieve a signal of 48khz, well above human hearing, capping out at 96khz. Why choose too use up computing power (taps) when, as far as I understand, their would be no audible gain? Again this is my understanding, please feel free if you are so inclined to inform me
If the DSP uses different samplingrates the filters has to be altered. So when using 1 samplerate in the DSP, only 1 set of filters are needed.
For digital sources there is not any gain in going to 192kHz exept not bandlimit 192kHz material (i think). At analog in it is the same reason of not bandlimiting the source.
To save computing power at lower frequencies, all except the highest frequency driver, you downsample to save taps. Often the HF over 3000 Hz need little prosessing so the penalty is not that big. What you can hear is another story. (For me 44,1 16 bit is enough
Regarding the resampling of 192khz. I dont know what is intended here, but some ASRC chips reclock all inputs and thereby introduce some signal prosessing on all input samplerates. This is done so that the DSP controls the samplerate and the filters don't alter characteristics with small changes in input samplerate.
For digital sources there is not any gain in going to 192kHz exept not bandlimit 192kHz material (i think). At analog in it is the same reason of not bandlimiting the source.
To save computing power at lower frequencies, all except the highest frequency driver, you downsample to save taps. Often the HF over 3000 Hz need little prosessing so the penalty is not that big. What you can hear is another story. (For me 44,1 16 bit is enough
Regarding the resampling of 192khz. I dont know what is intended here, but some ASRC chips reclock all inputs and thereby introduce some signal prosessing on all input samplerates. This is done so that the DSP controls the samplerate and the filters don't alter characteristics with small changes in input samplerate.
Last edited:
So as I anciently await the completion of this project to implement it in my system as an active 4 way crossover, or at least to start.
So at the risk of hi-jacking this thread, I propose a question. I am trying to understand why one would have the need for a word length of 192khz or better. If you can achieve a signal of 48khz, well above human hearing, capping out at 96khz. Why choose too use up computing power (taps) when, as far as I understand, their would be no audible gain? Again this is my understanding, please feel free if you are so inclined to inform me
I hear the difference. To me, 192kHz sounds better than lower sample rates. I don't want to debate the subject on this thread. It's been beat to death in countless other threads and forums
If the DSP uses different samplingrates the filters has to be altered. So when using 1 samplerate in the DSP, only 1 set of filters are needed.
For digital sources there is not any gain in going to 192kHz exept not bandlimit 192kHz material (i think). At analog in it is the same reason of not bandlimiting the source.
To save computing power at lower frequencies, all except the highest frequency driver, you downsample to save taps. Often the HF over 3000 Hz need little prosessing so the penalty is not that big. What you can hear is another story. (For me 44,1 16 bit is enough
Regarding the resampling of 192khz. I dont know what is intended here, but some ASRC chips reclock all inputs and thereby introduce some signal prosessing on all input samplerates. This is done so that the DSP controls the samplerate and the filters don't alter characteristics with small changes in input samplerate.
Thinking about your comment, 44.1/16 is good enough, I believe that is because most music that is listened to is a 44.1/ 16 recording. If you can find a recording that is captured at a higher rate say 24/96, and never down sampled, I believe you can her the difference. Higher dynamic range/ higher frincquncy response.
By "For me" I ment "For me personally". I don't hear much over 15k. And I also believe in masking. So maybe I hear single 18k signals but not in the presence of 10k signals with higher intensity.
I don't doubt that there are some lucky ones out there that has ears, physically out of the ordinary, that can hear 20k+
As to the 16 bit, you need more bits for prosessing, but 96 db is pretty high dynamic range before the DAC. I don't know any microphones with much better S/N. And when playing 120 dB, 25dB noise is hardly anoying.
On top of that you can noise shape so actual dunamic range is 100+ dB.
The difference I have heard between hires and normal records I think comes from different mastering of the material.
To be fear i haven't tested much on live "unmastered" material like this:
DPA Microphones :: stereo-recordings
But I am satisfied with the music I like and all the energy in that music is under 20k.
But as a final remark: I am not against having some margins so doubling or quadruple bandwith is not a bad idea when DSP resources are high and DSP is cheap.
I don't doubt that there are some lucky ones out there that has ears, physically out of the ordinary, that can hear 20k+
As to the 16 bit, you need more bits for prosessing, but 96 db is pretty high dynamic range before the DAC. I don't know any microphones with much better S/N. And when playing 120 dB, 25dB noise is hardly anoying.
On top of that you can noise shape so actual dunamic range is 100+ dB.
The difference I have heard between hires and normal records I think comes from different mastering of the material.
To be fear i haven't tested much on live "unmastered" material like this:
DPA Microphones :: stereo-recordings
But I am satisfied with the music I like and all the energy in that music is under 20k.
But as a final remark: I am not against having some margins so doubling or quadruple bandwith is not a bad idea when DSP resources are high and DSP is cheap.
Last edited:
Latency is only relevant when doing video (lip sync). Cant you wait to hear the music an other 100ms - are you in such a hurry ?
Latency that varies with frequency is an other thing. Maybe that what you ment? Because if the music starts 2 ms earlier for a 192k recording than for a 44k - what does it matter?
But if one starts to do manipulation in DSP it takes longer for a higher fs signal due to the computation effort. But again, whats the hurry - delay will not vary with freq while in the digital domain.
//
?Latency that varies with frequency is an other thing. Maybe that what you ment? Because if the music starts 2 ms earlier for a 192k recording than for a 44k - what does it matter?
But if one starts to do manipulation in DSP it takes longer for a higher fs signal due to the computation effort. But again, whats the hurry - delay will not vary with freq while in the digital domain.
//
Latency does not matter with video - modern DAWs have latency compensation, it's automated now.
Latency does matter with live sound and digital desks. Most of them run at 96KHz now with small buffer sizes to minimise the latency.
Latency does not matter at ALL with hifi playback, TNT you are right about that.
Latency does matter with live sound and digital desks. Most of them run at 96KHz now with small buffer sizes to minimise the latency.
Latency does not matter at ALL with hifi playback, TNT you are right about that.
And TV sound is complicated with long latency, then you can't take it directly from most TVs.
My comment was mainly aimed at the fact that this is a thread about a high end audio DSP that in most cases will have much more complex and steep filters than the anti aliasing filters in the DAC.
I also think most will use decimation with sharp filters to have higher frequency resolution in their FIR filters.
Many people seem to forget that frequency resolution of FIR filters is Samplerate/FIRlength
So for 1 Hz frequency resolution at 192k you need FIRlength of 192k. At 512Hz samplerate you need FIRlength of 512. And processing time is FIRlength squared.
Anyway 192k is standard in DAC and ADC nowadays.
Maybe there will be an option to set the ASRC to 48kHz output for the anti highres people?
Or we can decimate directly after the ASRC if we bother, as the frequency resolution over 3000Hz can be low in most cases.
My comment was mainly aimed at the fact that this is a thread about a high end audio DSP that in most cases will have much more complex and steep filters than the anti aliasing filters in the DAC.
I also think most will use decimation with sharp filters to have higher frequency resolution in their FIR filters.
Many people seem to forget that frequency resolution of FIR filters is Samplerate/FIRlength
So for 1 Hz frequency resolution at 192k you need FIRlength of 192k. At 512Hz samplerate you need FIRlength of 512. And processing time is FIRlength squared.
Anyway 192k is standard in DAC and ADC nowadays.
Maybe there will be an option to set the ASRC to 48kHz output for the anti highres people?
Or we can decimate directly after the ASRC if we bother, as the frequency resolution over 3000Hz can be low in most cases.
This is what I love about this place, without actually asking you can come away with what you seek. My gifts tend toward cabinetry, I can purchase drivers and build nice looking boxes to house them, done, easy. What's not so easy, is getting them to sound as good as they possibly can. Not my gifts but, learning to utilize DSP is the next step in this process, so if I ask a question that sounds a little naïve, so be it. Just working it out. There are something's that I have worked out, and that is there has to be a point of diminishing returns. And I know that stepping on someone's religion is sticking my hand into a hornet's nest, and I'm not referring to computation, but recording and play back. Anything above 24/96 is a waist of bandwidth. And isn't playback what this is all about. There I said it.
Last edited: