Shadorne - You have no clue. Filter will smooth-out the steps but will never remove few Hz modulation that was shown at 21Hz sampling rate. Al, please help me here or I'm going to kill myself.
Irvrobinson - I'm not talking about frequency range of our hearing but rather resolution of our hearing similar to number of shades of gray you can distinguish taking into consideration adverse conditions like ambient noise, system noise, THD, IMD etc.
I don't care anymore to defend myself from such attacks. You guys have no basic education in electronics and post nonsense just to keep arguing. Signing off. |
Mizuno - in order to avoid aliasing there should be no signal at 1/2 of the sampling frequency. In order to achieve it data has to be filtered out at 1/2 of sampling frequency in A/D processing.
Notice, that we are talking about preserving frequency information only (no aliases). Amplitude wise 16/44.1 will be very limited. Lets assume that you can hear 15kHz. Make picture of one full cycle of sinewave on a paper and try to place 3 points on it (reconstruct with 3 points only). You see the problem. Second problem is that filtering out info above 22.05kHz requires steep filters. Steep filters time shift different frequencies by different amount (uneven group delays) making inaccurate summing of harmonics. This will also screw-up step response (transients). Steep filters are not used in SACD recording making step response better. Of course master tapes are recorded in higher rate and re-sampled down but 96kHz playback will be still better than 44.1kHz (more points). 192kHz contains even more points but playback at 192kHz is not necessarily better than at 96kHz where THD of the most D/A ICs is the lowest (unless DAC uses extra info - downsampling). Resolution wise 24bit is better but most of converters are limited to about 20 bits anyway. Traditional converters are limited by tolerance of components to about 18 bits while Delta-Sigma are limited by timing errors to about 20 bits. One possible exception is Ring-DAC used by DCS (and previously licensed to ARCAM) that gets extra resolution by switching identical components of divider ladder in order to obtain more accurate average value. Some of the resolution will get buried in system noise that comes either from jitter (noise in time domain)or power amp's S/N. |
Irvrobinson - 20kHz frequency is reproduced accurately (no aliasing). As for the amplitude, theorem assumes infinite number of samples (of periodic signal). Because it is not the case, interpolation is done with Sinc functions but with constantly changing signal that is close to 1/2 of sampling frequency it is very coarse. More samples would be better IMHO.
As for oversampling in A/D process - even if you sample at 192kHz your filters have to get 96dB attenuation at 96kHz to be 16-bit perfect. Such Bessel filter would have to have perhaps 16 or so poles. Attenuation of 20kHz/-3dB 8 pole Bessel filter is only 50dB at 96kHz. Fortunately signals at 20kHz have very low amplitude so that might be OK.
I like 16/44 and agree that a lot can be improved in other areas. Jitter, being source of noise, is one of them. We learned to remove jitter by better (dual) Phase Lock Loops or asynchronous rate converters (upsampling) but there is still some jitter from less than perfect A/D processing that cannot be removed (common for older recordings). |
"capturing the frequency domain information at 20KHz, 44.1KHz sampling is completely sufficient to perfectly capture the sine waves"
Maybe sufficient for sinewaves but not for the music because it would call for brick wall filters that have very uneven group delays (non-linear phase if you prefer) and will cause wrong summing of harmonics. Such setup will be OK for single frequency reproduction but will be very unpleasant with music (dynamic signal).
Yes it is coarse because Nyquist-Shannon theorem requires infinite amount of terms (samples). Fixing it with sin(x)/x works poorly for short bursts around 1/2 of the sampling frequency. Sound of instruments producing continuous sound might be not affected (like flute) but anything with transients will sound wrong (piano, percussion instr. etc). Notice, that when people compare analog to 16/44 first thing they notice is different sound of the cymbals.
On the other hand, if you still think it is perfect system - enjoy. |
I was trying to show that 16/44 recording wouldn't be a perfect process and that's why it is done in 24/192 but downsampling to 16/44 also takes away quality.
Digital reproduction (as well as analog) have limitations. Filtering screws up transient response and 16 bit resolution is less than perfect.
Why it is difficult to hear difference thru Benchmark? Possibly because available hi-res is often poorly made (many complains about that) while our systems and rooms have shortcomings.
Power amp might be limiting factor but it isn't as bad as Irvrobinson calculated. First of all S/N or THD+N of the amp is usually shown at 1W and many amps are better than 96dB. In addition we don't listen at 1W . For instance if we take Rowland 625 amp's S/N specification of 95dB at 1W at 8 ohm it will be higher at the output power of 300W. We might look as well at residual output noise specified by Rowland that is 55uV at 20Hz-20kHz unweighted. Since output voltage at nominal power of 300W is around 50V it makes S/N=119dB. SACD reproduction is roughly equivalent to 20/96 requiring dynamic range of 120dB. D/A converters are also limited to 20 bits performance.
So to answer original question - increasing resolution might be beneficial up to about 20 bits assuming good recording/file, system and room. Increasing rate will be always beneficial to avoid serious shortcomings I mentioned before.
I settled at standard redbook reproduction not only for practical reasons but also because I cannot stand hiss and pops of analog playback that don't allow me to forget I'm not sitting "there" at the concert. |
Shadorne, agree about compression but it is simply who is driving the market. Release uncompressed piano recording (about 96dB dynamics) and a lot of people will complain that on their boom boxes or shelf speakers woofers are constantly buzzing. Hi-res has different clientele so they reduced compression a bit but it is still bad. Also, as you mentioned, they try to make average loudness as high as possible because to inexperienced customer it appears as higher quality recording especially with poorly resolving systems. |
What can I say - I posted example showing that quality amplifier is not a limiting factor. Why not to respond to that? If you think you can find any mistake in my reasoning please say so. Even for my own Rowland 102 (a class D amp) dynamic range is stated as 110dB while Rowland 301 is rated 120dB. Every Krell is at least 106dB unweighted (Evolution 900e is 113dB unweighted related to full power). You can search for a bad amp but the point was to show that the amp is not the limiting factor.
As for the Benchmark DAC1 again - If you cannot hear the difference then you can not, but please don't bring Nyquist into discussion since his theorem was intended toward stationary waveforms (infinite number of samples). Closer you get to Nyquist frequency the more samples you need to properly reconstruct original waveform - not possible to do for short high frequency sounds. That is the main reason so many people still stay with vinyl (unless you think they like convenience). |
Al,
Thank you. You brought very important point - quality of the recording engineering (on the top of compression issue that Shadorne mentioned). Average quality is not very high while some of the recordings I have are just incredibly good. Perhaps I'm arguing too much for the best case scenario while average quality of the recording was another reason for me to stay with 16/44 and Benchmark DAC1.
I checked recording as well - not available. |
Mapman wrote "you definitely want very good, younger ears"
Oh yes, but where can I get it? |
Irvrobinson - I assume that you buy properly sized amp for the speakers and the room. My amp is rated 150W at 6ohm and I am pretty sure I am getting peaks even larger than that (headroom). It corresponds to largest digital number coming from CD - meaning covers full dynamic range. If you listen at 1W then I agree that you have no chance to experience full dynamic range, not only because of the noise floor of the amp but more likely because of the ambient noise and threshold of our hearing.
To test if power amp is limiting factor is very simple - Just turn on power amp, set volume to zero and listen. Can you hear anything? I cannot - dead silent. If I cannot hear anything in very quiet room in my listening position why even bring numbers into discussion?
As for Nyquist - digital reproduction is decent from 16/44 media and, according to reviews, pretty good with SACD. I seriously doubt that they would release 24/192 master tapes to public. What is released right know as high resolution is often the same as 16/44 (I read article about it). SACD is a different story because it cannot be copied (pit width modulation) but it does not work with the server and selection is very limited. I settled at 16/44 for all the reasons I mentioned before but understand its limitations. I adjusted my gear accordingly with very forgiving Hyperion speakers. |
Al, I found video to show what happens when sampling just above Nyquist frequency. It might be possible to fix the output with sinc or other reconstruction functions but only if signal lasts a lot of cycles. If signal is short and disappears reconstruction will have huge error.
http://www.youtube.com/watch?v=Fy9dJgGCWZI |
Al, Huge errors applied to the highest harmonics only will result only in small sound change. There will be small difference in sound of cymbals and perhaps in ambiance.
I use 16/44 and like it, but try to be educated about it. That's all. |
Bob - Thanks for the link. I suspect that THD is a dominating factor at higher power. Noise issue itself is non-existent in my opinion because if I cannot hear anything in a silent room at full power (dead silent) I don't worry. Many amps with similar 80dB THD+N performance are showing -120dB noise floor on the other graphs. Also, small amount of noise helps to increase resolution - technique known as dithering widely used in photography.
I would be more concerned with THD and it doesn't look good.
I don't know what is relationship between THD and resolution but I suspect that resolution will still bring better sound. Another reason for that is quantization noise that is smaller at higher resolutions. DAC1 does very good job here by using sigma-delta converter that pushes quantization noise to higher bandwidth (oversampling).
I think that our hearing ability ends up slightly above 16-bit perhaps 18-20bits but I'm more concerned with sampling rate because low sampling rate in addition to phase shifts in steep low pass filters increases quantization noise (or size of square steps to make it simpler). |
