I mainly posted that due to the filter discussion and Lavry. It seemed so "old" so I wanted to bring the more current standard into the picture and what can be done now. I'll try to measure it. I can do 192ksps/24b but it is just an ordinary external sound card so... There is also a one cap LPF on the output on the actual product that sets in at appr. 350 khz.
//
We can all have a go at some FIR filter design... Use member "POS" excellent software found at:
rePhase - Official Site - Free FIR filtering tool
//
rePhase - Official Site - Free FIR filtering tool
//
Last edited:
Actually, it is all over but the shouting. The AES studied what is really needed .. 24/96 moves the inaccuracy to lower treble and doubling that takes you past 20Khz for accurate sound. So, that is all done. Now we just have emi/rfi and IM to deal with going into the next chassis. So, there are several ways to be sure HF isnt a problem.
Now, the continuous wave models and all CD design is based upon it seems maybe not the best way for music waveforms. BUT, 24/96 and beyond kinda makes it moot. Just keep the HF noise down, please.
IMHO
THx-RNMarsh
Now, the continuous wave models and all CD design is based upon it seems maybe not the best way for music waveforms. BUT, 24/96 and beyond kinda makes it moot. Just keep the HF noise down, please.
IMHO
THx-RNMarsh
Last edited:
Just to be clear - I indeed use it in my DAC. But the posted graphs where from the filter design software.
//
Could be. Sallen-Key filters are especially useless when the op-amp runs out of loop gain. MFB at least have a passive RC baked into the topology. The datasheet examples shouldn't be too bad in general, but most of those circuits rely on the downstream equipment to reject any common-mode on XLR outputs.
There's no real inaccuracy IMO, but even if you perceive it to be an issue, 24/96 has been around for just shy of 30 years now. TI released the PCM1702/4 in 1993 and I'm not even sure that was the first DAC IC to support 96 kHz sample rates including 8x oversampling.
Looking up that datasheet made me feel old 🙂.
Well, I disagree and thats part of my point. JC and many critical listeners and musicians have said so since CD arrived. Why 16/44 of CD is so fiercely defended here is strange. AES found it insufficient a long time ago. I just said you can hear it isnt like the real deal, especially as the frequency increases.
And, I decided for myself, with good reason, it was a lot due to insufficient sampling. This should not be new news to anyone. But, I guess LP and CD adherents just dont want to change and think 24/96+ is just a scam. ??
Anyway.... what was I doing ...?
-Richard
Last edited:
Do we now and then miss to indicate if we talk about the concept or a real world implementation? There are many ways to make poor down-sampling.
A parallel recording. One set of mic's, split into 2 feeds - 2 pcs of the same make/model A/D converter, one set to 24/44,1 (16/44?) and the other to 24/192. Playback these 2 files over the same system and make an evaluation.
Avoid aliasing and mirroring like the plague.
It's obvious that Mr Marsh has had better experience, somehow - for some reason (maybe not actual BW? (I say probably not)), with higher sample-rate material - we can't doubt that. It's his experience. Sad part for him the available music is limited. And at least a few of those downloads are just upsampled 44,1 as I understand it - scams. This is why there is a value in pursuing great quality for 16/44 hifi playback. Luckily I think it's possible.
//
I am not worried about s/h droop.
Re-examine the 22khz 500 point data graph, the fish looking one. Take that very dataset, and stuff it into the reconstruction block. Any level of oversampling you wish. Yes, you will get a pretty good looking 22 kHz sine, but what amplitude? The start of the data has almost no energy, so how can the reconstruction know what the steady state level is. 100 point in the level is still low.
The modulation envelope is a sine modulation, with a period of 880 points. Unless the output filter spans much more than that, it cannot accurately do the math.
How wide would an FFT window need to be to accurately portray the energy in the data? 16, 32, 64, 128, 256 all are too small to capture it repeatably and accurately as the output is starting time dependent.
Ps. 22050-22000= 50
22000/50 = 440
If you look carefully at the 500 sample plot, note that the full envelope has a modulation period of 440 samples. This is the beating between the sine and the sampling frequency. The closer the two, the longer the modulation envelope, and the wider the window has to be to see what's going on.
This is the accuracy vs sampling freq thing, the closer to nyquist, the longer the dataset needed.
To require extended windows in the lab to accurately capture the waveform energy, but yet assume an oversampling system can extract the accurate waveform in 1, 2,10,50 samples is folly.
Jn
Re-examine the 22khz 500 point data graph, the fish looking one. Take that very dataset, and stuff it into the reconstruction block. Any level of oversampling you wish. Yes, you will get a pretty good looking 22 kHz sine, but what amplitude? The start of the data has almost no energy, so how can the reconstruction know what the steady state level is. 100 point in the level is still low.
The modulation envelope is a sine modulation, with a period of 880 points. Unless the output filter spans much more than that, it cannot accurately do the math.
How wide would an FFT window need to be to accurately portray the energy in the data? 16, 32, 64, 128, 256 all are too small to capture it repeatably and accurately as the output is starting time dependent.
Ps. 22050-22000= 50
22000/50 = 440
If you look carefully at the 500 sample plot, note that the full envelope has a modulation period of 440 samples. This is the beating between the sine and the sampling frequency. The closer the two, the longer the modulation envelope, and the wider the window has to be to see what's going on.
This is the accuracy vs sampling freq thing, the closer to nyquist, the longer the dataset needed.
To require extended windows in the lab to accurately capture the waveform energy, but yet assume an oversampling system can extract the accurate waveform in 1, 2,10,50 samples is folly.
Jn
Last edited:
Ok, I thought "fold" would mean something like removed, so understand.
But, as a thought experiment; as we know, ideal sampling and D/A conversion (Fs=44.1 kHz) leads to the known ideal dirac pulse train which is represented by the spectral images at each multiple of the sampling frequency.
If zero stuffing alone would move the images to higher frequency, what will happen if we are doing more and more zero stuffing, shorten the accordingly the hold time and in the extreme case approximate (or even equal) the ideal dirac pulse train?
At that point the image spectra must all of a sudden return, because we know that there is/was a unique (one to one) relation between the dirac pulse train and its spectral representation (as shown in the first row of the first image below).
As a more practical example, Hans Polak's colleagues at Philips, messieurs Goedhart, van de Plassche and Stikvoort illustrated their four times oversampling system for the first Philips CD-player:
First row the ideal spectral representation of sampling with an idealised dirac pulse train at 44.1 Khz,
second row the resulting spectral representation after zero stuffing and applying the digital filter and finally
third row, the overlaying amplitude attenuation introduced by the NRZ (ZOH) of the "new" four times oversampled output.
Passband and stopband (beginning) of the digital filter used (compensation included for the remaining amplitude attenuation of the upper audio band frequencies,due to the NRZ and the analog third order Bessel of -3dB at 30 kHz) ):
They thought, that the digital filter should provide at least 50 dB attenuation at 22.05 kHz, realised as a linear phase filter.
As shown, the images were not moved by the zero-stuffing-process but attenuated by the digital low pass filter (therefore residue are still present due to the limited attenuation of the digital filter)
Last edited:
That would be another interesting test to see what is affected that way.
If you saw # 33811, we have significant artifacts shown from 20Khz to 1 MHz.
But then in the second CD photo, there is a lot of signal energy coming thru peaked at only 46KHz.
Seems few even care to see above 20KHz as no one is going to measure distortion up there on a CD/DAC player.
I made a nice little 5 pole flat GD filter at 75KHz (passive) and put it at output of CD player. But, it is too high freq with the noise coming in right after 22KHz. I'll redo it for a lower cutoff and see.
Would be good IMO to have BW standard be 44KHz and not 22Khz.
But still have to get rid of the HF. Your LC filter may be the most practical and better way.
THx-RNMarsh
Last edited:
I'd be perfectly happy with 24/96 becoming the standard, but there is no solid proof it's audibly superior, IMO. It makes life easier from an engineering standpoint and data is cheap, so why not?
Some musicians hate all digital, some hate CD, some hate DSD, some hate hi-res PCM. I'm not sure why we care what they think. Anyone standing in front of a stack of Marshalls or PA speakers on a regular basis without hearing protection is probably deaf AF. Even un-amplified, a lot of musicians suffer from hearing loss. My father has played Bb trumpet and soprano bugle (G trumpet) for 40+ years and HF hearing loss is not uncommon among his peers.
We all know this only goes in circles. John C and others will never accept digital. Mark and those that ascribe to his theories will tell you that it has to be high-res or HQPlayer DSD-512 to bypass the sigma-delta modulator of the DAC to sound good.
I often see claims in this thread that something must be audible because so many people on internet agree. They often ignore groups of people that feel the opposite of course. The public at large has had 30 years to digest "high-res" PCM and I would have to say it's been soundly rejected. People will line up to buy 4k and soon 8k LCD or OLED TVs, but the general public doesn't even know or care what sample rate their music is delivered in. Conveniently, these differences disappear in DBT as well.
You know, I do have an open mind. I suspect we both might be members of an elite group that once bought at least one LaserDisc. 🙂
I will be shocked if you never owned a LaserDisc player.
Last edited:
Exactly. For example, member midrange is a professional violinist with a London orchestra, his posts about his experience are interesting
Nice explanation;
All the way thru design choices were marginal. 50dB only. With a preamp and power amp gains of 50dB total.... that is not too cool. Only specs related to the CD player as a single item were tested and not the System affect.
This sort of thing has been a source of confounders for audiophiles and reviewers who have said something about the sound is not right about CD but standard measures fine. Thus leading to the great debates. And, may be different affect on different down stream gear. More confusion. But, they trust what they hear. EE ought to try to find out why instead of insisting everything is fine, its just you all imagination or group hypnosis etc.
Maybe some did listen and found 24/96 and higher was truely needed for accurate sound. And, we all lived happily ever after.
The End.
For now.
THx-RNMarsh
Last edited:
- Status
- Not open for further replies.