You said:
The simple response to your question is that perceived harshness can be caused by multiple factors. One cause that has been reported in certain cases has to the presence of ultrasonics, but that's not always a causal factor. Often it a combination of various factors that can cause a dac to sound "so harsh."
The simple response to your question is that perceived harshness can be caused by multiple factors. One cause that has been reported in certain cases has to the presence of ultrasonics, but that's not always a causal factor. Often it a combination of various factors that can cause a dac to sound "so harsh."
I would be careful to accept this thesisis as a fact.
The Gordon Holt test was done ages ago and Digital and Analog filters have much improved since
Against this theory is that nobody with 192/24 is complaining of harsh sound and neither are the NOSDAC lovers with huge amounts of mirrors.
Hans
I downloaded for free a handbook describing in details the PLL https://www.ti.com.cn/cn/lit/ml/snaa106c/snaa106c.pdf
It shows a phenomena that very low frequency filtering becomes bad for jitter. By this I decided to make it single 3 pole of higher quality than dual.
It shows a phenomena that very low frequency filtering becomes bad for jitter. By this I decided to make it single 3 pole of higher quality than dual.
This is with single PLL.
It multiplies by (16-24) times according to the DAC used and also multiplies by (8-32) according to bit rate 48-192kbit/s.
I have doubts about resetting the FIFO, it needs timing adjustments.
A special jitter eliminating PLL with unity frequency gain can be used, the low cost (199$) SI5317 in series with the actual one running at 27-36Mhz, for 16-24bit.
Attachments
Last edited:
One swallow does not make a summer.
And what about the 96/24 and 192/24 and even 384/24 recordings, not to speak about DSD with huge amounts of HF ?
I read quite often that they are supposed to produce superior sound.
In another thread I have used a 192/24 recording with cymbals, a very critical instrument to reproduce.
After having stripped the content resp. to 44.1/24 and 44.1/16, I have upsampled again to 192/24 and added non correlating content above 22.05Khz and below 16 bits to make it hard to find the master with an analyzer.
Guess what, nobody was able by listening to tell the unprocessed master from the other two.
The most convinced person of them all, who advocated 192/24 as far superior, even selected the 44.1/16.
And playing the above stripped 44.1/16 was not sounding different before and after upsampling to 192/24. Which is to be expected since no new information is added.
And also after adding uncorrelated content above 22.05K and below 16bits, there was absolutely no difference to be heard.
So as far as I am concerned, a proper 44.1/16 cannot be improved with higher bitrate or wordlength.
Some say that content above 20K make the sound harsh, others say they only buy 96K or higher because it sounds so much better.
I think the truth is somewhere in the middle, depending on your audio set.
Can your audio gear cope with HF without audible anomalies or not.
But for me, high res is just a commercial issue to bring new revenue into the world of content.
Hans
Many people like me have tons of 44.1k/16b accumulated since decades, my purpose is to reproduce at best. The higher sampling rates don't have much problems.
The question is that since the second generation of NOS CD players, it became obvious that the image frequencies should be cleared either by analog or digital filters, but why?. My approach to this problem is by addition and subtraction that I cancel the image and equalize the signal. This is electronics, I want to understand why I must cancel the ultrasounds image frequencies (24k to 30k), so that ears hear the music right? Only the understanding can say, down to what level it is sufficient to cancel out.
The question is that since the second generation of NOS CD players, it became obvious that the image frequencies should be cleared either by analog or digital filters, but why?. My approach to this problem is by addition and subtraction that I cancel the image and equalize the signal. This is electronics, I want to understand why I must cancel the ultrasounds image frequencies (24k to 30k), so that ears hear the music right? Only the understanding can say, down to what level it is sufficient to cancel out.
Last edited:
The problem of playing CDs such that they sound nearly perfect is a difficult one, but it has already been solved. The next problem is that it is not cheap to do well. Start with a good dac, say Topping D90 (or even better if one can afford it). Then play or rip the CDs using a PC. Buy the program 'HQ Player' to upsample and remodulate the 16/44 PCM to DSD256. The PC will need to be very powerful to run the most computation intensive algorithms in the program, but there are some algorithms that can run on almost any PC. I can run some of them with a 10-year old laptop.
Okay, now for the bad news: Topping D90 costs around $700. HQ Player costs around $250. An SOA PC costs several thousand $$$. You see the problem we have now, right? We need to find a way to bring the cost down.
There is some good news, however: There is a thread in the forum for a project that does 16/44 PCM to DSD256 processing in hardware at low cost. The bad news is the chips needed to build the project are not available right now because of the AKM plant fire.
Okay, now for the bad news: Topping D90 costs around $700. HQ Player costs around $250. An SOA PC costs several thousand $$$. You see the problem we have now, right? We need to find a way to bring the cost down.
There is some good news, however: There is a thread in the forum for a project that does 16/44 PCM to DSD256 processing in hardware at low cost. The bad news is the chips needed to build the project are not available right now because of the AKM plant fire.
Last edited:
Hayk,
Out of interest, could you give an example of a second generation NOS dac.
In what respect is it second generation.
You will have a sinc frequency envelope that has to be corrected either digitally or analogue.
When doing this in the analog domain, you will need a very complex filter to counteract the Sinc below 20Khz and some brickwall filter at 22.05Khz.
When doing this in the digital domain, you could just as well upsample to 192Khz, have no sinc correction problems and be left with a very relaxed analog filter.
So I have the feeling that the digital option has many advantages, but the Fir filter used after upsampling is obviously to blame for eventual loss in quality.
I can’t remember who it was, but a member on this forum has built his own very much longer Fir filter after upsampling and reported a significant increase in audio quality.
Hans
P.S. I’m sure you know, but upsampling is nothing else but inserting zeros in the signal stream.
A process that is fully transparant.
Out of interest, could you give an example of a second generation NOS dac.
In what respect is it second generation.
You will have a sinc frequency envelope that has to be corrected either digitally or analogue.
When doing this in the analog domain, you will need a very complex filter to counteract the Sinc below 20Khz and some brickwall filter at 22.05Khz.
When doing this in the digital domain, you could just as well upsample to 192Khz, have no sinc correction problems and be left with a very relaxed analog filter.
So I have the feeling that the digital option has many advantages, but the Fir filter used after upsampling is obviously to blame for eventual loss in quality.
I can’t remember who it was, but a member on this forum has built his own very much longer Fir filter after upsampling and reported a significant increase in audio quality.
Hans
P.S. I’m sure you know, but upsampling is nothing else but inserting zeros in the signal stream.
A process that is fully transparant.
Last edited:
There is two problems and not one to be treated due to sampling. The first, is the step character of the DAC output. This is treated by interpolation, either by adding zeros between two samples and average the result, or modulate the data by sigma-delta (DSD is the same) and get values between two samples. The second problem is the modulation. The Nyquist frequency modulates in amplitude and the frequencies above ½Nyquist get a beat due to presence of their images. This is treated by filtering out image frequencies by FIR digital filter or elliptic analog after the DAC output, both introducing resonances. The single treatment that repairs both problems, is the sinc interpolation. It can interpolate all the values missed by the sampler and the beat disappears. Two new problems appear by sinc interpolation, the number of samples dealt should be very high and the result has significant pre and post echo resonances.
My way of using radio technic of SSB for image elimination and level equalization along with cubic spline interpolation, gave superior results.
My way of using radio technic of SSB for image elimination and level equalization along with cubic spline interpolation, gave superior results.
I see two descriptions of the same problem in your post, one description in the time domain and one description in the frequency domain.
In any case, what is still missing is the effect of the anti-aliasing filter. No-one would sample audio signals without first band-limiting them with an anti-aliasing filter. When this is a nearly ideal filter that just removes everything above 22050 Hz and leaves everything below it unscathed, you get long ultrasonic pre- and post-ringing, because an ideal filter has a sinc-shaped impulse response. After sampling and reconstruction with an ideal sinc-shaped impulse response, you get the same waveform as at the output of the anti-aliasing filter. For example, with a 20 kHz tone burst and 44.1 kHz sample rate, an ideal anti-aliasing filter would change the waveform of the tone burst into the red waveform in the attachment, sampling would change it into the green waveform and reconstruction would change it back into the red waveform.
It is possible to reduce the ringing if you are willing to accept treble loss. For example, with a relatively smooth low-pass filter of which the stop band starts at a lower frequency than the transition bands of any of the other filters in the signal chain. Peter Garde dubbed that an apodization filter.
In any case, what is still missing is the effect of the anti-aliasing filter. No-one would sample audio signals without first band-limiting them with an anti-aliasing filter. When this is a nearly ideal filter that just removes everything above 22050 Hz and leaves everything below it unscathed, you get long ultrasonic pre- and post-ringing, because an ideal filter has a sinc-shaped impulse response. After sampling and reconstruction with an ideal sinc-shaped impulse response, you get the same waveform as at the output of the anti-aliasing filter. For example, with a 20 kHz tone burst and 44.1 kHz sample rate, an ideal anti-aliasing filter would change the waveform of the tone burst into the red waveform in the attachment, sampling would change it into the green waveform and reconstruction would change it back into the red waveform.
It is possible to reduce the ringing if you are willing to accept treble loss. For example, with a relatively smooth low-pass filter of which the stop band starts at a lower frequency than the transition bands of any of the other filters in the signal chain. Peter Garde dubbed that an apodization filter.
Attachments
If you sample a single 20khz sine wave, it doesn't need any anti-aliasing filter. You sample this frequency at 44.1khz and it generates 24khz image frequency. To eliminate this image 24khz ultrasound, to this day what I knew, is to filter by digital or analog means. The third way is mine, which is superior to what you are proposing. All I want to understand, why we need to get ride of this ultrasound to get it sounding right? I firmly believe the waves on the output of the DAC above 11khz are spaced mostly at 22khz interval with less and less defect as it approaches the Nyquist. May be the ear doesn't demodulate as the FFT to get the below Nyquist components unless the ultasounds once filtered it is becoming demodulated. This how I explain at this point of my understanding. This problem is there since 40 years and only one month that I am dealing with. Corona period looks to last at least a year more, so I have lot of time to think about solutions.
Last edited:
What I want to say that even the signal before sampling is exempt of ultrasound, the sampler does create them, that is required from the DAC to filter them. My mission is not to record but playback only what the CD producer have sold to me. If it is recorded with resonances, I can do nothing about it, but I need to avoid adding resonances because I want to make it better.
That's just the point: assuming that during recording, the signal was somehow band-limited to below 22050 Hz, which it had to be to prevent aliasing, reconstructing the signal by passing it through an ideal low-pass at 22050 Hz is not going to add any pre- or post-ringing. It just reconstructs the band-limited waveform that was sampled.
Nonetheless, you could try to remove ultrasonic pre- and/or post-ringing that was already there at the anti-aliasing filter's output by using an apodizing filter as reconstruction filter - which will reduce treble.
Nonetheless, you could try to remove ultrasonic pre- and/or post-ringing that was already there at the anti-aliasing filter's output by using an apodizing filter as reconstruction filter - which will reduce treble.
Last edited:
I don't understand why you are talking about ant-aliasing, it concerns the recording. The pre and post resonances are not ultrasonic but the signals frequency perfectly audible, this is due to the nature of the digital filter.
What I am convinced that people are filtering since 40 years the image ultrasonic frequencies not knowing for why. I couldn't find anywhere test to determine how low should the image frequencies be. In late 80's when CD players became popular, I tested these low cost player at the vendors, the best sounding I found was the German Dual with Z filter written. Most of these models used third party low pass modules.
What I am convinced that people are filtering since 40 years the image ultrasonic frequencies not knowing for why. I couldn't find anywhere test to determine how low should the image frequencies be. In late 80's when CD players became popular, I tested these low cost player at the vendors, the best sounding I found was the German Dual with Z filter written. Most of these models used third party low pass modules.
Sorry, but you are wrong this time how upsampling works.
When upsampling, zero's are inserted, that's correct.
The spectrum you will get then is that the original spectrum is still mirrored around multiples of 44.1, just like before, which is not to be surprised because no info has been added.
The thing that has happened though, is that when feeding this data stream to a DAC, the Sinc envelope has its zero's shifted to a much higher frequency, depending on how many zero's were inserted.
Let's say 3 zero's were inserted, the first Sinc zero will now be at 176Khz, giving an attenuation of a negligible -0,18dB@20Khz instead of the previous -3,17dB you had before upsampling.
But this is still without FIR filter.
This Fir filter can effectively remove all content between 22.05Khz and 88.2Khz. So it's no interpolation but simply LP filtering.
That will make that your analogue filter now has 7 octaves at its disposal to go from 22.05Khz to 154.35Khz to filter HF, very easy to implement.
Thus the FIR filter is to blame when quality of audio was affected, not the zero stuffing, so it must be long enough to prevent damage.
Hans
P.S. Can you still give me an example of a second generation NOS Dac that you referred to?
You try to reproduce what was recorded, don't you? Then it's important to know that the recorded signal had a bandwidth limitation.
Time domain: the impulse response of an ideal low pass filter at 22050 Hz is sin(2*pi*22050 Hz*t)/(2*pi*22050 Hz*t) for t not equal to 0, 1 for t = 0. Most people would regard 22050 Hz ringing as ultrasonic.
Frequency domain: all an ideal low-pass filter at 22050 Hz does is to remove everything above 22050 Hz. Hence, the ringing is due to the absence of the components above 22050 Hz. Most people would regard these as ultrasonic.
They need to have a sin(2 pi fN t)/(2 pi fN t)-shaped impulse response for perfect anti-alias filtering and perfect reconstruction, and that response is symmetrical. fN is the Nyquist frequency, which is half the sample frequency. When the bandwidth at the recording side was limited with some other type of filter, you can still reconstruct what came out of that filter by using a filter with a sin(2 pi fN t)/(2 pi fN t)-shaped impulse response on the DAC side.